1 //===- BugReporter.cpp - Generate PathDiagnostics for bugs ----------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file defines BugReporter, a utility class for generating 10 // PathDiagnostics. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" 15 #include "clang/AST/Decl.h" 16 #include "clang/AST/DeclBase.h" 17 #include "clang/AST/DeclObjC.h" 18 #include "clang/AST/Expr.h" 19 #include "clang/AST/ExprCXX.h" 20 #include "clang/AST/ParentMap.h" 21 #include "clang/AST/Stmt.h" 22 #include "clang/AST/StmtCXX.h" 23 #include "clang/AST/StmtObjC.h" 24 #include "clang/Analysis/AnalysisDeclContext.h" 25 #include "clang/Analysis/CFG.h" 26 #include "clang/Analysis/CFGStmtMap.h" 27 #include "clang/Analysis/PathDiagnostic.h" 28 #include "clang/Analysis/ProgramPoint.h" 29 #include "clang/Basic/LLVM.h" 30 #include "clang/Basic/SourceLocation.h" 31 #include "clang/Basic/SourceManager.h" 32 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h" 33 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h" 34 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 35 #include "clang/StaticAnalyzer/Core/Checker.h" 36 #include "clang/StaticAnalyzer/Core/CheckerManager.h" 37 #include "clang/StaticAnalyzer/Core/CheckerRegistryData.h" 38 #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" 39 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" 40 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h" 41 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" 42 #include "clang/StaticAnalyzer/Core/PathSensitive/SMTConv.h" 43 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" 44 #include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" 45 #include "llvm/ADT/ArrayRef.h" 46 #include "llvm/ADT/DenseMap.h" 47 #include "llvm/ADT/DenseSet.h" 48 #include "llvm/ADT/FoldingSet.h" 49 #include "llvm/ADT/STLExtras.h" 50 #include "llvm/ADT/SmallPtrSet.h" 51 #include "llvm/ADT/SmallString.h" 52 #include "llvm/ADT/SmallVector.h" 53 #include "llvm/ADT/Statistic.h" 54 #include "llvm/ADT/StringExtras.h" 55 #include "llvm/ADT/StringRef.h" 56 #include "llvm/ADT/iterator_range.h" 57 #include "llvm/Support/Casting.h" 58 #include "llvm/Support/Compiler.h" 59 #include "llvm/Support/ErrorHandling.h" 60 #include "llvm/Support/MemoryBuffer.h" 61 #include "llvm/Support/raw_ostream.h" 62 #include <algorithm> 63 #include <cassert> 64 #include <cstddef> 65 #include <iterator> 66 #include <memory> 67 #include <optional> 68 #include <queue> 69 #include <string> 70 #include <tuple> 71 #include <utility> 72 #include <vector> 73 74 using namespace clang; 75 using namespace ento; 76 using namespace llvm; 77 78 #define DEBUG_TYPE "BugReporter" 79 80 STATISTIC(MaxBugClassSize, 81 "The maximum number of bug reports in the same equivalence class"); 82 STATISTIC(MaxValidBugClassSize, 83 "The maximum number of bug reports in the same equivalence class " 84 "where at least one report is valid (not suppressed)"); 85 86 BugReporterVisitor::~BugReporterVisitor() = default; 87 88 void BugReporterContext::anchor() {} 89 90 //===----------------------------------------------------------------------===// 91 // PathDiagnosticBuilder and its associated routines and helper objects. 92 //===----------------------------------------------------------------------===// 93 94 namespace { 95 96 /// A (CallPiece, node assiciated with its CallEnter) pair. 97 using CallWithEntry = 98 std::pair<PathDiagnosticCallPiece *, const ExplodedNode *>; 99 using CallWithEntryStack = SmallVector<CallWithEntry, 6>; 100 101 /// Map from each node to the diagnostic pieces visitors emit for them. 102 using VisitorsDiagnosticsTy = 103 llvm::DenseMap<const ExplodedNode *, std::vector<PathDiagnosticPieceRef>>; 104 105 /// A map from PathDiagnosticPiece to the LocationContext of the inlined 106 /// function call it represents. 107 using LocationContextMap = 108 llvm::DenseMap<const PathPieces *, const LocationContext *>; 109 110 /// A helper class that contains everything needed to construct a 111 /// PathDiagnostic object. It does no much more then providing convenient 112 /// getters and some well placed asserts for extra security. 113 class PathDiagnosticConstruct { 114 /// The consumer we're constructing the bug report for. 115 const PathDiagnosticConsumer *Consumer; 116 /// Our current position in the bug path, which is owned by 117 /// PathDiagnosticBuilder. 118 const ExplodedNode *CurrentNode; 119 /// A mapping from parts of the bug path (for example, a function call, which 120 /// would span backwards from a CallExit to a CallEnter with the nodes in 121 /// between them) with the location contexts it is associated with. 122 LocationContextMap LCM; 123 const SourceManager &SM; 124 125 public: 126 /// We keep stack of calls to functions as we're ascending the bug path. 127 /// TODO: PathDiagnostic has a stack doing the same thing, shouldn't we use 128 /// that instead? 129 CallWithEntryStack CallStack; 130 /// The bug report we're constructing. For ease of use, this field is kept 131 /// public, though some "shortcut" getters are provided for commonly used 132 /// methods of PathDiagnostic. 133 std::unique_ptr<PathDiagnostic> PD; 134 135 public: 136 PathDiagnosticConstruct(const PathDiagnosticConsumer *PDC, 137 const ExplodedNode *ErrorNode, 138 const PathSensitiveBugReport *R); 139 140 /// \returns the location context associated with the current position in the 141 /// bug path. 142 const LocationContext *getCurrLocationContext() const { 143 assert(CurrentNode && "Already reached the root!"); 144 return CurrentNode->getLocationContext(); 145 } 146 147 /// Same as getCurrLocationContext (they should always return the same 148 /// location context), but works after reaching the root of the bug path as 149 /// well. 150 const LocationContext *getLocationContextForActivePath() const { 151 return LCM.find(&PD->getActivePath())->getSecond(); 152 } 153 154 const ExplodedNode *getCurrentNode() const { return CurrentNode; } 155 156 /// Steps the current node to its predecessor. 157 /// \returns whether we reached the root of the bug path. 158 bool ascendToPrevNode() { 159 CurrentNode = CurrentNode->getFirstPred(); 160 return static_cast<bool>(CurrentNode); 161 } 162 163 const ParentMap &getParentMap() const { 164 return getCurrLocationContext()->getParentMap(); 165 } 166 167 const SourceManager &getSourceManager() const { return SM; } 168 169 const Stmt *getParent(const Stmt *S) const { 170 return getParentMap().getParent(S); 171 } 172 173 void updateLocCtxMap(const PathPieces *Path, const LocationContext *LC) { 174 assert(Path && LC); 175 LCM[Path] = LC; 176 } 177 178 const LocationContext *getLocationContextFor(const PathPieces *Path) const { 179 assert(LCM.count(Path) && 180 "Failed to find the context associated with these pieces!"); 181 return LCM.find(Path)->getSecond(); 182 } 183 184 bool isInLocCtxMap(const PathPieces *Path) const { return LCM.count(Path); } 185 186 PathPieces &getActivePath() { return PD->getActivePath(); } 187 PathPieces &getMutablePieces() { return PD->getMutablePieces(); } 188 189 bool shouldAddPathEdges() const { return Consumer->shouldAddPathEdges(); } 190 bool shouldAddControlNotes() const { 191 return Consumer->shouldAddControlNotes(); 192 } 193 bool shouldGenerateDiagnostics() const { 194 return Consumer->shouldGenerateDiagnostics(); 195 } 196 bool supportsLogicalOpControlFlow() const { 197 return Consumer->supportsLogicalOpControlFlow(); 198 } 199 }; 200 201 /// Contains every contextual information needed for constructing a 202 /// PathDiagnostic object for a given bug report. This class and its fields are 203 /// immutable, and passes a BugReportConstruct object around during the 204 /// construction. 205 class PathDiagnosticBuilder : public BugReporterContext { 206 /// A linear path from the error node to the root. 207 std::unique_ptr<const ExplodedGraph> BugPath; 208 /// The bug report we're describing. Visitors create their diagnostics with 209 /// them being the last entities being able to modify it (for example, 210 /// changing interestingness here would cause inconsistencies as to how this 211 /// file and visitors construct diagnostics), hence its const. 212 const PathSensitiveBugReport *R; 213 /// The leaf of the bug path. This isn't the same as the bug reports error 214 /// node, which refers to the *original* graph, not the bug path. 215 const ExplodedNode *const ErrorNode; 216 /// The diagnostic pieces visitors emitted, which is expected to be collected 217 /// by the time this builder is constructed. 218 std::unique_ptr<const VisitorsDiagnosticsTy> VisitorsDiagnostics; 219 220 public: 221 /// Find a non-invalidated report for a given equivalence class, and returns 222 /// a PathDiagnosticBuilder able to construct bug reports for different 223 /// consumers. Returns std::nullopt if no valid report is found. 224 static std::optional<PathDiagnosticBuilder> 225 findValidReport(ArrayRef<PathSensitiveBugReport *> &bugReports, 226 PathSensitiveBugReporter &Reporter); 227 228 PathDiagnosticBuilder( 229 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath, 230 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode, 231 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics); 232 233 /// This function is responsible for generating diagnostic pieces that are 234 /// *not* provided by bug report visitors. 235 /// These diagnostics may differ depending on the consumer's settings, 236 /// and are therefore constructed separately for each consumer. 237 /// 238 /// There are two path diagnostics generation modes: with adding edges (used 239 /// for plists) and without (used for HTML and text). When edges are added, 240 /// the path is modified to insert artificially generated edges. 241 /// Otherwise, more detailed diagnostics is emitted for block edges, 242 /// explaining the transitions in words. 243 std::unique_ptr<PathDiagnostic> 244 generate(const PathDiagnosticConsumer *PDC) const; 245 246 private: 247 void updateStackPiecesWithMessage(PathDiagnosticPieceRef P, 248 const CallWithEntryStack &CallStack) const; 249 void generatePathDiagnosticsForNode(PathDiagnosticConstruct &C, 250 PathDiagnosticLocation &PrevLoc) const; 251 252 void generateMinimalDiagForBlockEdge(PathDiagnosticConstruct &C, 253 BlockEdge BE) const; 254 255 PathDiagnosticPieceRef 256 generateDiagForGotoOP(const PathDiagnosticConstruct &C, const Stmt *S, 257 PathDiagnosticLocation &Start) const; 258 259 PathDiagnosticPieceRef 260 generateDiagForSwitchOP(const PathDiagnosticConstruct &C, const CFGBlock *Dst, 261 PathDiagnosticLocation &Start) const; 262 263 PathDiagnosticPieceRef 264 generateDiagForBinaryOP(const PathDiagnosticConstruct &C, const Stmt *T, 265 const CFGBlock *Src, const CFGBlock *DstC) const; 266 267 PathDiagnosticLocation 268 ExecutionContinues(const PathDiagnosticConstruct &C) const; 269 270 PathDiagnosticLocation 271 ExecutionContinues(llvm::raw_string_ostream &os, 272 const PathDiagnosticConstruct &C) const; 273 274 const PathSensitiveBugReport *getBugReport() const { return R; } 275 }; 276 277 } // namespace 278 279 //===----------------------------------------------------------------------===// 280 // Base implementation of stack hint generators. 281 //===----------------------------------------------------------------------===// 282 283 StackHintGenerator::~StackHintGenerator() = default; 284 285 std::string StackHintGeneratorForSymbol::getMessage(const ExplodedNode *N){ 286 if (!N) 287 return getMessageForSymbolNotFound(); 288 289 ProgramPoint P = N->getLocation(); 290 CallExitEnd CExit = P.castAs<CallExitEnd>(); 291 292 // FIXME: Use CallEvent to abstract this over all calls. 293 const Stmt *CallSite = CExit.getCalleeContext()->getCallSite(); 294 const auto *CE = dyn_cast_or_null<CallExpr>(CallSite); 295 if (!CE) 296 return {}; 297 298 // Check if one of the parameters are set to the interesting symbol. 299 for (auto [Idx, ArgExpr] : llvm::enumerate(CE->arguments())) { 300 SVal SV = N->getSVal(ArgExpr); 301 302 // Check if the variable corresponding to the symbol is passed by value. 303 SymbolRef AS = SV.getAsLocSymbol(); 304 if (AS == Sym) { 305 return getMessageForArg(ArgExpr, Idx); 306 } 307 308 // Check if the parameter is a pointer to the symbol. 309 if (std::optional<loc::MemRegionVal> Reg = SV.getAs<loc::MemRegionVal>()) { 310 // Do not attempt to dereference void*. 311 if (ArgExpr->getType()->isVoidPointerType()) 312 continue; 313 SVal PSV = N->getState()->getSVal(Reg->getRegion()); 314 SymbolRef AS = PSV.getAsLocSymbol(); 315 if (AS == Sym) { 316 return getMessageForArg(ArgExpr, Idx); 317 } 318 } 319 } 320 321 // Check if we are returning the interesting symbol. 322 SVal SV = N->getSVal(CE); 323 SymbolRef RetSym = SV.getAsLocSymbol(); 324 if (RetSym == Sym) { 325 return getMessageForReturn(CE); 326 } 327 328 return getMessageForSymbolNotFound(); 329 } 330 331 std::string StackHintGeneratorForSymbol::getMessageForArg(const Expr *ArgE, 332 unsigned ArgIndex) { 333 // Printed parameters start at 1, not 0. 334 ++ArgIndex; 335 336 return (llvm::Twine(Msg) + " via " + std::to_string(ArgIndex) + 337 llvm::getOrdinalSuffix(ArgIndex) + " parameter").str(); 338 } 339 340 //===----------------------------------------------------------------------===// 341 // Diagnostic cleanup. 342 //===----------------------------------------------------------------------===// 343 344 static PathDiagnosticEventPiece * 345 eventsDescribeSameCondition(PathDiagnosticEventPiece *X, 346 PathDiagnosticEventPiece *Y) { 347 // Prefer diagnostics that come from ConditionBRVisitor over 348 // those that came from TrackConstraintBRVisitor, 349 // unless the one from ConditionBRVisitor is 350 // its generic fallback diagnostic. 351 const void *tagPreferred = ConditionBRVisitor::getTag(); 352 const void *tagLesser = TrackConstraintBRVisitor::getTag(); 353 354 if (X->getLocation() != Y->getLocation()) 355 return nullptr; 356 357 if (X->getTag() == tagPreferred && Y->getTag() == tagLesser) 358 return ConditionBRVisitor::isPieceMessageGeneric(X) ? Y : X; 359 360 if (Y->getTag() == tagPreferred && X->getTag() == tagLesser) 361 return ConditionBRVisitor::isPieceMessageGeneric(Y) ? X : Y; 362 363 return nullptr; 364 } 365 366 /// An optimization pass over PathPieces that removes redundant diagnostics 367 /// generated by both ConditionBRVisitor and TrackConstraintBRVisitor. Both 368 /// BugReporterVisitors use different methods to generate diagnostics, with 369 /// one capable of emitting diagnostics in some cases but not in others. This 370 /// can lead to redundant diagnostic pieces at the same point in a path. 371 static void removeRedundantMsgs(PathPieces &path) { 372 unsigned N = path.size(); 373 if (N < 2) 374 return; 375 // NOTE: this loop intentionally is not using an iterator. Instead, we 376 // are streaming the path and modifying it in place. This is done by 377 // grabbing the front, processing it, and if we decide to keep it append 378 // it to the end of the path. The entire path is processed in this way. 379 for (unsigned i = 0; i < N; ++i) { 380 auto piece = std::move(path.front()); 381 path.pop_front(); 382 383 switch (piece->getKind()) { 384 case PathDiagnosticPiece::Call: 385 removeRedundantMsgs(cast<PathDiagnosticCallPiece>(*piece).path); 386 break; 387 case PathDiagnosticPiece::Macro: 388 removeRedundantMsgs(cast<PathDiagnosticMacroPiece>(*piece).subPieces); 389 break; 390 case PathDiagnosticPiece::Event: { 391 if (i == N-1) 392 break; 393 394 if (auto *nextEvent = 395 dyn_cast<PathDiagnosticEventPiece>(path.front().get())) { 396 auto *event = cast<PathDiagnosticEventPiece>(piece.get()); 397 // Check to see if we should keep one of the two pieces. If we 398 // come up with a preference, record which piece to keep, and consume 399 // another piece from the path. 400 if (auto *pieceToKeep = 401 eventsDescribeSameCondition(event, nextEvent)) { 402 piece = std::move(pieceToKeep == event ? piece : path.front()); 403 path.pop_front(); 404 ++i; 405 } 406 } 407 break; 408 } 409 case PathDiagnosticPiece::ControlFlow: 410 case PathDiagnosticPiece::Note: 411 case PathDiagnosticPiece::PopUp: 412 break; 413 } 414 path.push_back(std::move(piece)); 415 } 416 } 417 418 /// Recursively scan through a path and prune out calls and macros pieces 419 /// that aren't needed. Return true if afterwards the path contains 420 /// "interesting stuff" which means it shouldn't be pruned from the parent path. 421 static bool removeUnneededCalls(const PathDiagnosticConstruct &C, 422 PathPieces &pieces, 423 const PathSensitiveBugReport *R, 424 bool IsInteresting = false) { 425 bool containsSomethingInteresting = IsInteresting; 426 const unsigned N = pieces.size(); 427 428 for (unsigned i = 0 ; i < N ; ++i) { 429 // Remove the front piece from the path. If it is still something we 430 // want to keep once we are done, we will push it back on the end. 431 auto piece = std::move(pieces.front()); 432 pieces.pop_front(); 433 434 switch (piece->getKind()) { 435 case PathDiagnosticPiece::Call: { 436 auto &call = cast<PathDiagnosticCallPiece>(*piece); 437 // Check if the location context is interesting. 438 if (!removeUnneededCalls( 439 C, call.path, R, 440 R->isInteresting(C.getLocationContextFor(&call.path)))) 441 continue; 442 443 containsSomethingInteresting = true; 444 break; 445 } 446 case PathDiagnosticPiece::Macro: { 447 auto ¯o = cast<PathDiagnosticMacroPiece>(*piece); 448 if (!removeUnneededCalls(C, macro.subPieces, R, IsInteresting)) 449 continue; 450 containsSomethingInteresting = true; 451 break; 452 } 453 case PathDiagnosticPiece::Event: { 454 auto &event = cast<PathDiagnosticEventPiece>(*piece); 455 456 // We never throw away an event, but we do throw it away wholesale 457 // as part of a path if we throw the entire path away. 458 containsSomethingInteresting |= !event.isPrunable(); 459 break; 460 } 461 case PathDiagnosticPiece::ControlFlow: 462 case PathDiagnosticPiece::Note: 463 case PathDiagnosticPiece::PopUp: 464 break; 465 } 466 467 pieces.push_back(std::move(piece)); 468 } 469 470 return containsSomethingInteresting; 471 } 472 473 /// Same logic as above to remove extra pieces. 474 static void removePopUpNotes(PathPieces &Path) { 475 for (unsigned int i = 0; i < Path.size(); ++i) { 476 auto Piece = std::move(Path.front()); 477 Path.pop_front(); 478 if (!isa<PathDiagnosticPopUpPiece>(*Piece)) 479 Path.push_back(std::move(Piece)); 480 } 481 } 482 483 /// Returns true if the given decl has been implicitly given a body, either by 484 /// the analyzer or by the compiler proper. 485 static bool hasImplicitBody(const Decl *D) { 486 assert(D); 487 return D->isImplicit() || !D->hasBody(); 488 } 489 490 /// Recursively scan through a path and make sure that all call pieces have 491 /// valid locations. 492 static void 493 adjustCallLocations(PathPieces &Pieces, 494 PathDiagnosticLocation *LastCallLocation = nullptr) { 495 for (const auto &I : Pieces) { 496 auto *Call = dyn_cast<PathDiagnosticCallPiece>(I.get()); 497 498 if (!Call) 499 continue; 500 501 if (LastCallLocation) { 502 bool CallerIsImplicit = hasImplicitBody(Call->getCaller()); 503 if (CallerIsImplicit || !Call->callEnter.asLocation().isValid()) 504 Call->callEnter = *LastCallLocation; 505 if (CallerIsImplicit || !Call->callReturn.asLocation().isValid()) 506 Call->callReturn = *LastCallLocation; 507 } 508 509 // Recursively clean out the subclass. Keep this call around if 510 // it contains any informative diagnostics. 511 PathDiagnosticLocation *ThisCallLocation; 512 if (Call->callEnterWithin.asLocation().isValid() && 513 !hasImplicitBody(Call->getCallee())) 514 ThisCallLocation = &Call->callEnterWithin; 515 else 516 ThisCallLocation = &Call->callEnter; 517 518 assert(ThisCallLocation && "Outermost call has an invalid location"); 519 adjustCallLocations(Call->path, ThisCallLocation); 520 } 521 } 522 523 /// Remove edges in and out of C++ default initializer expressions. These are 524 /// for fields that have in-class initializers, as opposed to being initialized 525 /// explicitly in a constructor or braced list. 526 static void removeEdgesToDefaultInitializers(PathPieces &Pieces) { 527 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { 528 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get())) 529 removeEdgesToDefaultInitializers(C->path); 530 531 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get())) 532 removeEdgesToDefaultInitializers(M->subPieces); 533 534 if (auto *CF = dyn_cast<PathDiagnosticControlFlowPiece>(I->get())) { 535 const Stmt *Start = CF->getStartLocation().asStmt(); 536 const Stmt *End = CF->getEndLocation().asStmt(); 537 if (isa_and_nonnull<CXXDefaultInitExpr>(Start)) { 538 I = Pieces.erase(I); 539 continue; 540 } else if (isa_and_nonnull<CXXDefaultInitExpr>(End)) { 541 PathPieces::iterator Next = std::next(I); 542 if (Next != E) { 543 if (auto *NextCF = 544 dyn_cast<PathDiagnosticControlFlowPiece>(Next->get())) { 545 NextCF->setStartLocation(CF->getStartLocation()); 546 } 547 } 548 I = Pieces.erase(I); 549 continue; 550 } 551 } 552 553 I++; 554 } 555 } 556 557 /// Remove all pieces with invalid locations as these cannot be serialized. 558 /// We might have pieces with invalid locations as a result of inlining Body 559 /// Farm generated functions. 560 static void removePiecesWithInvalidLocations(PathPieces &Pieces) { 561 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { 562 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get())) 563 removePiecesWithInvalidLocations(C->path); 564 565 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get())) 566 removePiecesWithInvalidLocations(M->subPieces); 567 568 if (!(*I)->getLocation().isValid() || 569 !(*I)->getLocation().asLocation().isValid()) { 570 I = Pieces.erase(I); 571 continue; 572 } 573 I++; 574 } 575 } 576 577 PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues( 578 const PathDiagnosticConstruct &C) const { 579 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics()) 580 return PathDiagnosticLocation(S, getSourceManager(), 581 C.getCurrLocationContext()); 582 583 return PathDiagnosticLocation::createDeclEnd(C.getCurrLocationContext(), 584 getSourceManager()); 585 } 586 587 PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues( 588 llvm::raw_string_ostream &os, const PathDiagnosticConstruct &C) const { 589 // Slow, but probably doesn't matter. 590 if (os.str().empty()) 591 os << ' '; 592 593 const PathDiagnosticLocation &Loc = ExecutionContinues(C); 594 595 if (Loc.asStmt()) 596 os << "Execution continues on line " 597 << getSourceManager().getExpansionLineNumber(Loc.asLocation()) 598 << '.'; 599 else { 600 os << "Execution jumps to the end of the "; 601 const Decl *D = C.getCurrLocationContext()->getDecl(); 602 if (isa<ObjCMethodDecl>(D)) 603 os << "method"; 604 else if (isa<FunctionDecl>(D)) 605 os << "function"; 606 else { 607 assert(isa<BlockDecl>(D)); 608 os << "anonymous block"; 609 } 610 os << '.'; 611 } 612 613 return Loc; 614 } 615 616 static const Stmt *getEnclosingParent(const Stmt *S, const ParentMap &PM) { 617 if (isa<Expr>(S) && PM.isConsumedExpr(cast<Expr>(S))) 618 return PM.getParentIgnoreParens(S); 619 620 const Stmt *Parent = PM.getParentIgnoreParens(S); 621 if (!Parent) 622 return nullptr; 623 624 switch (Parent->getStmtClass()) { 625 case Stmt::ForStmtClass: 626 case Stmt::DoStmtClass: 627 case Stmt::WhileStmtClass: 628 case Stmt::ObjCForCollectionStmtClass: 629 case Stmt::CXXForRangeStmtClass: 630 return Parent; 631 default: 632 break; 633 } 634 635 return nullptr; 636 } 637 638 static PathDiagnosticLocation 639 getEnclosingStmtLocation(const Stmt *S, const LocationContext *LC, 640 bool allowNestedContexts = false) { 641 if (!S) 642 return {}; 643 644 const SourceManager &SMgr = LC->getDecl()->getASTContext().getSourceManager(); 645 646 while (const Stmt *Parent = getEnclosingParent(S, LC->getParentMap())) { 647 switch (Parent->getStmtClass()) { 648 case Stmt::BinaryOperatorClass: { 649 const auto *B = cast<BinaryOperator>(Parent); 650 if (B->isLogicalOp()) 651 return PathDiagnosticLocation(allowNestedContexts ? B : S, SMgr, LC); 652 break; 653 } 654 case Stmt::CompoundStmtClass: 655 case Stmt::StmtExprClass: 656 return PathDiagnosticLocation(S, SMgr, LC); 657 case Stmt::ChooseExprClass: 658 // Similar to '?' if we are referring to condition, just have the edge 659 // point to the entire choose expression. 660 if (allowNestedContexts || cast<ChooseExpr>(Parent)->getCond() == S) 661 return PathDiagnosticLocation(Parent, SMgr, LC); 662 else 663 return PathDiagnosticLocation(S, SMgr, LC); 664 case Stmt::BinaryConditionalOperatorClass: 665 case Stmt::ConditionalOperatorClass: 666 // For '?', if we are referring to condition, just have the edge point 667 // to the entire '?' expression. 668 if (allowNestedContexts || 669 cast<AbstractConditionalOperator>(Parent)->getCond() == S) 670 return PathDiagnosticLocation(Parent, SMgr, LC); 671 else 672 return PathDiagnosticLocation(S, SMgr, LC); 673 case Stmt::CXXForRangeStmtClass: 674 if (cast<CXXForRangeStmt>(Parent)->getBody() == S) 675 return PathDiagnosticLocation(S, SMgr, LC); 676 break; 677 case Stmt::DoStmtClass: 678 return PathDiagnosticLocation(S, SMgr, LC); 679 case Stmt::ForStmtClass: 680 if (cast<ForStmt>(Parent)->getBody() == S) 681 return PathDiagnosticLocation(S, SMgr, LC); 682 break; 683 case Stmt::IfStmtClass: 684 if (cast<IfStmt>(Parent)->getCond() != S) 685 return PathDiagnosticLocation(S, SMgr, LC); 686 break; 687 case Stmt::ObjCForCollectionStmtClass: 688 if (cast<ObjCForCollectionStmt>(Parent)->getBody() == S) 689 return PathDiagnosticLocation(S, SMgr, LC); 690 break; 691 case Stmt::WhileStmtClass: 692 if (cast<WhileStmt>(Parent)->getCond() != S) 693 return PathDiagnosticLocation(S, SMgr, LC); 694 break; 695 default: 696 break; 697 } 698 699 S = Parent; 700 } 701 702 assert(S && "Cannot have null Stmt for PathDiagnosticLocation"); 703 704 return PathDiagnosticLocation(S, SMgr, LC); 705 } 706 707 //===----------------------------------------------------------------------===// 708 // "Minimal" path diagnostic generation algorithm. 709 //===----------------------------------------------------------------------===// 710 711 /// If the piece contains a special message, add it to all the call pieces on 712 /// the active stack. For example, my_malloc allocated memory, so MallocChecker 713 /// will construct an event at the call to malloc(), and add a stack hint that 714 /// an allocated memory was returned. We'll use this hint to construct a message 715 /// when returning from the call to my_malloc 716 /// 717 /// void *my_malloc() { return malloc(sizeof(int)); } 718 /// void fishy() { 719 /// void *ptr = my_malloc(); // returned allocated memory 720 /// } // leak 721 void PathDiagnosticBuilder::updateStackPiecesWithMessage( 722 PathDiagnosticPieceRef P, const CallWithEntryStack &CallStack) const { 723 if (R->hasCallStackHint(P)) 724 for (const auto &I : CallStack) { 725 PathDiagnosticCallPiece *CP = I.first; 726 const ExplodedNode *N = I.second; 727 std::string stackMsg = R->getCallStackMessage(P, N); 728 729 // The last message on the path to final bug is the most important 730 // one. Since we traverse the path backwards, do not add the message 731 // if one has been previously added. 732 if (!CP->hasCallStackMessage()) 733 CP->setCallStackMessage(stackMsg); 734 } 735 } 736 737 static void CompactMacroExpandedPieces(PathPieces &path, 738 const SourceManager& SM); 739 740 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForSwitchOP( 741 const PathDiagnosticConstruct &C, const CFGBlock *Dst, 742 PathDiagnosticLocation &Start) const { 743 744 const SourceManager &SM = getSourceManager(); 745 // Figure out what case arm we took. 746 std::string sbuf; 747 llvm::raw_string_ostream os(sbuf); 748 PathDiagnosticLocation End; 749 750 if (const Stmt *S = Dst->getLabel()) { 751 End = PathDiagnosticLocation(S, SM, C.getCurrLocationContext()); 752 753 switch (S->getStmtClass()) { 754 default: 755 os << "No cases match in the switch statement. " 756 "Control jumps to line " 757 << End.asLocation().getExpansionLineNumber(); 758 break; 759 case Stmt::DefaultStmtClass: 760 os << "Control jumps to the 'default' case at line " 761 << End.asLocation().getExpansionLineNumber(); 762 break; 763 764 case Stmt::CaseStmtClass: { 765 os << "Control jumps to 'case "; 766 const auto *Case = cast<CaseStmt>(S); 767 const Expr *LHS = Case->getLHS()->IgnoreParenImpCasts(); 768 769 // Determine if it is an enum. 770 bool GetRawInt = true; 771 772 if (const auto *DR = dyn_cast<DeclRefExpr>(LHS)) { 773 // FIXME: Maybe this should be an assertion. Are there cases 774 // were it is not an EnumConstantDecl? 775 const auto *D = dyn_cast<EnumConstantDecl>(DR->getDecl()); 776 777 if (D) { 778 GetRawInt = false; 779 os << *D; 780 } 781 } 782 783 if (GetRawInt) 784 os << LHS->EvaluateKnownConstInt(getASTContext()); 785 786 os << ":' at line " << End.asLocation().getExpansionLineNumber(); 787 break; 788 } 789 } 790 } else { 791 os << "'Default' branch taken. "; 792 End = ExecutionContinues(os, C); 793 } 794 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 795 os.str()); 796 } 797 798 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForGotoOP( 799 const PathDiagnosticConstruct &C, const Stmt *S, 800 PathDiagnosticLocation &Start) const { 801 std::string sbuf; 802 llvm::raw_string_ostream os(sbuf); 803 const PathDiagnosticLocation &End = 804 getEnclosingStmtLocation(S, C.getCurrLocationContext()); 805 os << "Control jumps to line " << End.asLocation().getExpansionLineNumber(); 806 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str()); 807 } 808 809 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForBinaryOP( 810 const PathDiagnosticConstruct &C, const Stmt *T, const CFGBlock *Src, 811 const CFGBlock *Dst) const { 812 813 const SourceManager &SM = getSourceManager(); 814 815 const auto *B = cast<BinaryOperator>(T); 816 std::string sbuf; 817 llvm::raw_string_ostream os(sbuf); 818 os << "Left side of '"; 819 PathDiagnosticLocation Start, End; 820 821 if (B->getOpcode() == BO_LAnd) { 822 os << "&&" 823 << "' is "; 824 825 if (*(Src->succ_begin() + 1) == Dst) { 826 os << "false"; 827 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 828 Start = 829 PathDiagnosticLocation::createOperatorLoc(B, SM); 830 } else { 831 os << "true"; 832 Start = 833 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 834 End = ExecutionContinues(C); 835 } 836 } else { 837 assert(B->getOpcode() == BO_LOr); 838 os << "||" 839 << "' is "; 840 841 if (*(Src->succ_begin() + 1) == Dst) { 842 os << "false"; 843 Start = 844 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 845 End = ExecutionContinues(C); 846 } else { 847 os << "true"; 848 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 849 Start = 850 PathDiagnosticLocation::createOperatorLoc(B, SM); 851 } 852 } 853 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 854 os.str()); 855 } 856 857 void PathDiagnosticBuilder::generateMinimalDiagForBlockEdge( 858 PathDiagnosticConstruct &C, BlockEdge BE) const { 859 const SourceManager &SM = getSourceManager(); 860 const LocationContext *LC = C.getCurrLocationContext(); 861 const CFGBlock *Src = BE.getSrc(); 862 const CFGBlock *Dst = BE.getDst(); 863 const Stmt *T = Src->getTerminatorStmt(); 864 if (!T) 865 return; 866 867 auto Start = PathDiagnosticLocation::createBegin(T, SM, LC); 868 switch (T->getStmtClass()) { 869 default: 870 break; 871 872 case Stmt::GotoStmtClass: 873 case Stmt::IndirectGotoStmtClass: { 874 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics()) 875 C.getActivePath().push_front(generateDiagForGotoOP(C, S, Start)); 876 break; 877 } 878 879 case Stmt::SwitchStmtClass: { 880 C.getActivePath().push_front(generateDiagForSwitchOP(C, Dst, Start)); 881 break; 882 } 883 884 case Stmt::BreakStmtClass: 885 case Stmt::ContinueStmtClass: { 886 std::string sbuf; 887 llvm::raw_string_ostream os(sbuf); 888 PathDiagnosticLocation End = ExecutionContinues(os, C); 889 C.getActivePath().push_front( 890 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str())); 891 break; 892 } 893 894 // Determine control-flow for ternary '?'. 895 case Stmt::BinaryConditionalOperatorClass: 896 case Stmt::ConditionalOperatorClass: { 897 std::string sbuf; 898 llvm::raw_string_ostream os(sbuf); 899 os << "'?' condition is "; 900 901 if (*(Src->succ_begin() + 1) == Dst) 902 os << "false"; 903 else 904 os << "true"; 905 906 PathDiagnosticLocation End = ExecutionContinues(C); 907 908 if (const Stmt *S = End.asStmt()) 909 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 910 911 C.getActivePath().push_front( 912 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str())); 913 break; 914 } 915 916 // Determine control-flow for short-circuited '&&' and '||'. 917 case Stmt::BinaryOperatorClass: { 918 if (!C.supportsLogicalOpControlFlow()) 919 break; 920 921 C.getActivePath().push_front(generateDiagForBinaryOP(C, T, Src, Dst)); 922 break; 923 } 924 925 case Stmt::DoStmtClass: 926 if (*(Src->succ_begin()) == Dst) { 927 std::string sbuf; 928 llvm::raw_string_ostream os(sbuf); 929 930 os << "Loop condition is true. "; 931 PathDiagnosticLocation End = ExecutionContinues(os, C); 932 933 if (const Stmt *S = End.asStmt()) 934 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 935 936 C.getActivePath().push_front( 937 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 938 os.str())); 939 } else { 940 PathDiagnosticLocation End = ExecutionContinues(C); 941 942 if (const Stmt *S = End.asStmt()) 943 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 944 945 C.getActivePath().push_front( 946 std::make_shared<PathDiagnosticControlFlowPiece>( 947 Start, End, "Loop condition is false. Exiting loop")); 948 } 949 break; 950 951 case Stmt::WhileStmtClass: 952 case Stmt::ForStmtClass: 953 if (*(Src->succ_begin() + 1) == Dst) { 954 std::string sbuf; 955 llvm::raw_string_ostream os(sbuf); 956 957 os << "Loop condition is false. "; 958 PathDiagnosticLocation End = ExecutionContinues(os, C); 959 if (const Stmt *S = End.asStmt()) 960 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 961 962 C.getActivePath().push_front( 963 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 964 os.str())); 965 } else { 966 PathDiagnosticLocation End = ExecutionContinues(C); 967 if (const Stmt *S = End.asStmt()) 968 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 969 970 C.getActivePath().push_front( 971 std::make_shared<PathDiagnosticControlFlowPiece>( 972 Start, End, "Loop condition is true. Entering loop body")); 973 } 974 975 break; 976 977 case Stmt::IfStmtClass: { 978 PathDiagnosticLocation End = ExecutionContinues(C); 979 980 if (const Stmt *S = End.asStmt()) 981 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 982 983 if (*(Src->succ_begin() + 1) == Dst) 984 C.getActivePath().push_front( 985 std::make_shared<PathDiagnosticControlFlowPiece>( 986 Start, End, "Taking false branch")); 987 else 988 C.getActivePath().push_front( 989 std::make_shared<PathDiagnosticControlFlowPiece>( 990 Start, End, "Taking true branch")); 991 992 break; 993 } 994 } 995 } 996 997 //===----------------------------------------------------------------------===// 998 // Functions for determining if a loop was executed 0 times. 999 //===----------------------------------------------------------------------===// 1000 1001 static bool isLoop(const Stmt *Term) { 1002 switch (Term->getStmtClass()) { 1003 case Stmt::ForStmtClass: 1004 case Stmt::WhileStmtClass: 1005 case Stmt::ObjCForCollectionStmtClass: 1006 case Stmt::CXXForRangeStmtClass: 1007 return true; 1008 default: 1009 // Note that we intentionally do not include do..while here. 1010 return false; 1011 } 1012 } 1013 1014 static bool isJumpToFalseBranch(const BlockEdge *BE) { 1015 const CFGBlock *Src = BE->getSrc(); 1016 assert(Src->succ_size() == 2); 1017 return (*(Src->succ_begin()+1) == BE->getDst()); 1018 } 1019 1020 static bool isContainedByStmt(const ParentMap &PM, const Stmt *S, 1021 const Stmt *SubS) { 1022 while (SubS) { 1023 if (SubS == S) 1024 return true; 1025 SubS = PM.getParent(SubS); 1026 } 1027 return false; 1028 } 1029 1030 static const Stmt *getStmtBeforeCond(const ParentMap &PM, const Stmt *Term, 1031 const ExplodedNode *N) { 1032 while (N) { 1033 std::optional<StmtPoint> SP = N->getLocation().getAs<StmtPoint>(); 1034 if (SP) { 1035 const Stmt *S = SP->getStmt(); 1036 if (!isContainedByStmt(PM, Term, S)) 1037 return S; 1038 } 1039 N = N->getFirstPred(); 1040 } 1041 return nullptr; 1042 } 1043 1044 static bool isInLoopBody(const ParentMap &PM, const Stmt *S, const Stmt *Term) { 1045 const Stmt *LoopBody = nullptr; 1046 switch (Term->getStmtClass()) { 1047 case Stmt::CXXForRangeStmtClass: { 1048 const auto *FR = cast<CXXForRangeStmt>(Term); 1049 if (isContainedByStmt(PM, FR->getInc(), S)) 1050 return true; 1051 if (isContainedByStmt(PM, FR->getLoopVarStmt(), S)) 1052 return true; 1053 LoopBody = FR->getBody(); 1054 break; 1055 } 1056 case Stmt::ForStmtClass: { 1057 const auto *FS = cast<ForStmt>(Term); 1058 if (isContainedByStmt(PM, FS->getInc(), S)) 1059 return true; 1060 LoopBody = FS->getBody(); 1061 break; 1062 } 1063 case Stmt::ObjCForCollectionStmtClass: { 1064 const auto *FC = cast<ObjCForCollectionStmt>(Term); 1065 LoopBody = FC->getBody(); 1066 break; 1067 } 1068 case Stmt::WhileStmtClass: 1069 LoopBody = cast<WhileStmt>(Term)->getBody(); 1070 break; 1071 default: 1072 return false; 1073 } 1074 return isContainedByStmt(PM, LoopBody, S); 1075 } 1076 1077 /// Adds a sanitized control-flow diagnostic edge to a path. 1078 static void addEdgeToPath(PathPieces &path, 1079 PathDiagnosticLocation &PrevLoc, 1080 PathDiagnosticLocation NewLoc) { 1081 if (!NewLoc.isValid()) 1082 return; 1083 1084 SourceLocation NewLocL = NewLoc.asLocation(); 1085 if (NewLocL.isInvalid()) 1086 return; 1087 1088 if (!PrevLoc.isValid() || !PrevLoc.asLocation().isValid()) { 1089 PrevLoc = NewLoc; 1090 return; 1091 } 1092 1093 // Ignore self-edges, which occur when there are multiple nodes at the same 1094 // statement. 1095 if (NewLoc.asStmt() && NewLoc.asStmt() == PrevLoc.asStmt()) 1096 return; 1097 1098 path.push_front( 1099 std::make_shared<PathDiagnosticControlFlowPiece>(NewLoc, PrevLoc)); 1100 PrevLoc = NewLoc; 1101 } 1102 1103 /// A customized wrapper for CFGBlock::getTerminatorCondition() 1104 /// which returns the element for ObjCForCollectionStmts. 1105 static const Stmt *getTerminatorCondition(const CFGBlock *B) { 1106 const Stmt *S = B->getTerminatorCondition(); 1107 if (const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(S)) 1108 return FS->getElement(); 1109 return S; 1110 } 1111 1112 constexpr llvm::StringLiteral StrEnteringLoop = "Entering loop body"; 1113 constexpr llvm::StringLiteral StrLoopBodyZero = "Loop body executed 0 times"; 1114 constexpr llvm::StringLiteral StrLoopRangeEmpty = 1115 "Loop body skipped when range is empty"; 1116 constexpr llvm::StringLiteral StrLoopCollectionEmpty = 1117 "Loop body skipped when collection is empty"; 1118 1119 static std::unique_ptr<FilesToLineNumsMap> 1120 findExecutedLines(const SourceManager &SM, const ExplodedNode *N); 1121 1122 void PathDiagnosticBuilder::generatePathDiagnosticsForNode( 1123 PathDiagnosticConstruct &C, PathDiagnosticLocation &PrevLoc) const { 1124 ProgramPoint P = C.getCurrentNode()->getLocation(); 1125 const SourceManager &SM = getSourceManager(); 1126 1127 // Have we encountered an entrance to a call? It may be 1128 // the case that we have not encountered a matching 1129 // call exit before this point. This means that the path 1130 // terminated within the call itself. 1131 if (auto CE = P.getAs<CallEnter>()) { 1132 1133 if (C.shouldAddPathEdges()) { 1134 // Add an edge to the start of the function. 1135 const StackFrameContext *CalleeLC = CE->getCalleeContext(); 1136 const Decl *D = CalleeLC->getDecl(); 1137 // Add the edge only when the callee has body. We jump to the beginning 1138 // of the *declaration*, however we expect it to be followed by the 1139 // body. This isn't the case for autosynthesized property accessors in 1140 // Objective-C. No need for a similar extra check for CallExit points 1141 // because the exit edge comes from a statement (i.e. return), 1142 // not from declaration. 1143 if (D->hasBody()) 1144 addEdgeToPath(C.getActivePath(), PrevLoc, 1145 PathDiagnosticLocation::createBegin(D, SM)); 1146 } 1147 1148 // Did we visit an entire call? 1149 bool VisitedEntireCall = C.PD->isWithinCall(); 1150 C.PD->popActivePath(); 1151 1152 PathDiagnosticCallPiece *Call; 1153 if (VisitedEntireCall) { 1154 Call = cast<PathDiagnosticCallPiece>(C.getActivePath().front().get()); 1155 } else { 1156 // The path terminated within a nested location context, create a new 1157 // call piece to encapsulate the rest of the path pieces. 1158 const Decl *Caller = CE->getLocationContext()->getDecl(); 1159 Call = PathDiagnosticCallPiece::construct(C.getActivePath(), Caller); 1160 assert(C.getActivePath().size() == 1 && 1161 C.getActivePath().front().get() == Call); 1162 1163 // Since we just transferred the path over to the call piece, reset the 1164 // mapping of the active path to the current location context. 1165 assert(C.isInLocCtxMap(&C.getActivePath()) && 1166 "When we ascend to a previously unvisited call, the active path's " 1167 "address shouldn't change, but rather should be compacted into " 1168 "a single CallEvent!"); 1169 C.updateLocCtxMap(&C.getActivePath(), C.getCurrLocationContext()); 1170 1171 // Record the location context mapping for the path within the call. 1172 assert(!C.isInLocCtxMap(&Call->path) && 1173 "When we ascend to a previously unvisited call, this must be the " 1174 "first time we encounter the caller context!"); 1175 C.updateLocCtxMap(&Call->path, CE->getCalleeContext()); 1176 } 1177 Call->setCallee(*CE, SM); 1178 1179 // Update the previous location in the active path. 1180 PrevLoc = Call->getLocation(); 1181 1182 if (!C.CallStack.empty()) { 1183 assert(C.CallStack.back().first == Call); 1184 C.CallStack.pop_back(); 1185 } 1186 return; 1187 } 1188 1189 assert(C.getCurrLocationContext() == C.getLocationContextForActivePath() && 1190 "The current position in the bug path is out of sync with the " 1191 "location context associated with the active path!"); 1192 1193 // Have we encountered an exit from a function call? 1194 if (std::optional<CallExitEnd> CE = P.getAs<CallExitEnd>()) { 1195 1196 // We are descending into a call (backwards). Construct 1197 // a new call piece to contain the path pieces for that call. 1198 auto Call = PathDiagnosticCallPiece::construct(*CE, SM); 1199 // Record the mapping from call piece to LocationContext. 1200 assert(!C.isInLocCtxMap(&Call->path) && 1201 "We just entered a call, this must've been the first time we " 1202 "encounter its context!"); 1203 C.updateLocCtxMap(&Call->path, CE->getCalleeContext()); 1204 1205 if (C.shouldAddPathEdges()) { 1206 // Add the edge to the return site. 1207 addEdgeToPath(C.getActivePath(), PrevLoc, Call->callReturn); 1208 PrevLoc.invalidate(); 1209 } 1210 1211 auto *P = Call.get(); 1212 C.getActivePath().push_front(std::move(Call)); 1213 1214 // Make the contents of the call the active path for now. 1215 C.PD->pushActivePath(&P->path); 1216 C.CallStack.push_back(CallWithEntry(P, C.getCurrentNode())); 1217 return; 1218 } 1219 1220 if (auto PS = P.getAs<PostStmt>()) { 1221 if (!C.shouldAddPathEdges()) 1222 return; 1223 1224 // Add an edge. If this is an ObjCForCollectionStmt do 1225 // not add an edge here as it appears in the CFG both 1226 // as a terminator and as a terminator condition. 1227 if (!isa<ObjCForCollectionStmt>(PS->getStmt())) { 1228 PathDiagnosticLocation L = 1229 PathDiagnosticLocation(PS->getStmt(), SM, C.getCurrLocationContext()); 1230 addEdgeToPath(C.getActivePath(), PrevLoc, L); 1231 } 1232 1233 } else if (auto BE = P.getAs<BlockEdge>()) { 1234 1235 if (C.shouldAddControlNotes()) { 1236 generateMinimalDiagForBlockEdge(C, *BE); 1237 } 1238 1239 if (!C.shouldAddPathEdges()) { 1240 return; 1241 } 1242 1243 // Are we jumping to the head of a loop? Add a special diagnostic. 1244 if (const Stmt *Loop = BE->getSrc()->getLoopTarget()) { 1245 PathDiagnosticLocation L(Loop, SM, C.getCurrLocationContext()); 1246 const Stmt *Body = nullptr; 1247 1248 if (const auto *FS = dyn_cast<ForStmt>(Loop)) 1249 Body = FS->getBody(); 1250 else if (const auto *WS = dyn_cast<WhileStmt>(Loop)) 1251 Body = WS->getBody(); 1252 else if (const auto *OFS = dyn_cast<ObjCForCollectionStmt>(Loop)) { 1253 Body = OFS->getBody(); 1254 } else if (const auto *FRS = dyn_cast<CXXForRangeStmt>(Loop)) { 1255 Body = FRS->getBody(); 1256 } 1257 // do-while statements are explicitly excluded here 1258 1259 auto p = std::make_shared<PathDiagnosticEventPiece>( 1260 L, "Looping back to the head of the loop"); 1261 p->setPrunable(true); 1262 1263 addEdgeToPath(C.getActivePath(), PrevLoc, p->getLocation()); 1264 // We might've added a very similar control node already 1265 if (!C.shouldAddControlNotes()) { 1266 C.getActivePath().push_front(std::move(p)); 1267 } 1268 1269 if (const auto *CS = dyn_cast_or_null<CompoundStmt>(Body)) { 1270 addEdgeToPath(C.getActivePath(), PrevLoc, 1271 PathDiagnosticLocation::createEndBrace(CS, SM)); 1272 } 1273 } 1274 1275 const CFGBlock *BSrc = BE->getSrc(); 1276 const ParentMap &PM = C.getParentMap(); 1277 1278 if (const Stmt *Term = BSrc->getTerminatorStmt()) { 1279 // Are we jumping past the loop body without ever executing the 1280 // loop (because the condition was false)? 1281 if (isLoop(Term)) { 1282 const Stmt *TermCond = getTerminatorCondition(BSrc); 1283 bool IsInLoopBody = isInLoopBody( 1284 PM, getStmtBeforeCond(PM, TermCond, C.getCurrentNode()), Term); 1285 1286 StringRef str; 1287 1288 if (isJumpToFalseBranch(&*BE)) { 1289 if (!IsInLoopBody) { 1290 if (isa<ObjCForCollectionStmt>(Term)) { 1291 str = StrLoopCollectionEmpty; 1292 } else if (isa<CXXForRangeStmt>(Term)) { 1293 str = StrLoopRangeEmpty; 1294 } else { 1295 str = StrLoopBodyZero; 1296 } 1297 } 1298 } else { 1299 str = StrEnteringLoop; 1300 } 1301 1302 if (!str.empty()) { 1303 PathDiagnosticLocation L(TermCond ? TermCond : Term, SM, 1304 C.getCurrLocationContext()); 1305 auto PE = std::make_shared<PathDiagnosticEventPiece>(L, str); 1306 PE->setPrunable(true); 1307 addEdgeToPath(C.getActivePath(), PrevLoc, PE->getLocation()); 1308 1309 // We might've added a very similar control node already 1310 if (!C.shouldAddControlNotes()) { 1311 C.getActivePath().push_front(std::move(PE)); 1312 } 1313 } 1314 } else if (isa<BreakStmt, ContinueStmt, GotoStmt>(Term)) { 1315 PathDiagnosticLocation L(Term, SM, C.getCurrLocationContext()); 1316 addEdgeToPath(C.getActivePath(), PrevLoc, L); 1317 } 1318 } 1319 } 1320 } 1321 1322 static std::unique_ptr<PathDiagnostic> 1323 generateDiagnosticForBasicReport(const BasicBugReport *R) { 1324 const BugType &BT = R->getBugType(); 1325 return std::make_unique<PathDiagnostic>( 1326 BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(), 1327 R->getDescription(), R->getShortDescription(/*UseFallback=*/false), 1328 BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(), 1329 std::make_unique<FilesToLineNumsMap>()); 1330 } 1331 1332 static std::unique_ptr<PathDiagnostic> 1333 generateEmptyDiagnosticForReport(const PathSensitiveBugReport *R, 1334 const SourceManager &SM) { 1335 const BugType &BT = R->getBugType(); 1336 return std::make_unique<PathDiagnostic>( 1337 BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(), 1338 R->getDescription(), R->getShortDescription(/*UseFallback=*/false), 1339 BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(), 1340 findExecutedLines(SM, R->getErrorNode())); 1341 } 1342 1343 static const Stmt *getStmtParent(const Stmt *S, const ParentMap &PM) { 1344 if (!S) 1345 return nullptr; 1346 1347 while (true) { 1348 S = PM.getParentIgnoreParens(S); 1349 1350 if (!S) 1351 break; 1352 1353 if (isa<FullExpr, CXXBindTemporaryExpr, SubstNonTypeTemplateParmExpr>(S)) 1354 continue; 1355 1356 break; 1357 } 1358 1359 return S; 1360 } 1361 1362 static bool isConditionForTerminator(const Stmt *S, const Stmt *Cond) { 1363 switch (S->getStmtClass()) { 1364 case Stmt::BinaryOperatorClass: { 1365 const auto *BO = cast<BinaryOperator>(S); 1366 if (!BO->isLogicalOp()) 1367 return false; 1368 return BO->getLHS() == Cond || BO->getRHS() == Cond; 1369 } 1370 case Stmt::IfStmtClass: 1371 return cast<IfStmt>(S)->getCond() == Cond; 1372 case Stmt::ForStmtClass: 1373 return cast<ForStmt>(S)->getCond() == Cond; 1374 case Stmt::WhileStmtClass: 1375 return cast<WhileStmt>(S)->getCond() == Cond; 1376 case Stmt::DoStmtClass: 1377 return cast<DoStmt>(S)->getCond() == Cond; 1378 case Stmt::ChooseExprClass: 1379 return cast<ChooseExpr>(S)->getCond() == Cond; 1380 case Stmt::IndirectGotoStmtClass: 1381 return cast<IndirectGotoStmt>(S)->getTarget() == Cond; 1382 case Stmt::SwitchStmtClass: 1383 return cast<SwitchStmt>(S)->getCond() == Cond; 1384 case Stmt::BinaryConditionalOperatorClass: 1385 return cast<BinaryConditionalOperator>(S)->getCond() == Cond; 1386 case Stmt::ConditionalOperatorClass: { 1387 const auto *CO = cast<ConditionalOperator>(S); 1388 return CO->getCond() == Cond || 1389 CO->getLHS() == Cond || 1390 CO->getRHS() == Cond; 1391 } 1392 case Stmt::ObjCForCollectionStmtClass: 1393 return cast<ObjCForCollectionStmt>(S)->getElement() == Cond; 1394 case Stmt::CXXForRangeStmtClass: { 1395 const auto *FRS = cast<CXXForRangeStmt>(S); 1396 return FRS->getCond() == Cond || FRS->getRangeInit() == Cond; 1397 } 1398 default: 1399 return false; 1400 } 1401 } 1402 1403 static bool isIncrementOrInitInForLoop(const Stmt *S, const Stmt *FL) { 1404 if (const auto *FS = dyn_cast<ForStmt>(FL)) 1405 return FS->getInc() == S || FS->getInit() == S; 1406 if (const auto *FRS = dyn_cast<CXXForRangeStmt>(FL)) 1407 return FRS->getInc() == S || FRS->getRangeStmt() == S || 1408 FRS->getLoopVarStmt() || FRS->getRangeInit() == S; 1409 return false; 1410 } 1411 1412 using OptimizedCallsSet = llvm::DenseSet<const PathDiagnosticCallPiece *>; 1413 1414 /// Adds synthetic edges from top-level statements to their subexpressions. 1415 /// 1416 /// This avoids a "swoosh" effect, where an edge from a top-level statement A 1417 /// points to a sub-expression B.1 that's not at the start of B. In these cases, 1418 /// we'd like to see an edge from A to B, then another one from B to B.1. 1419 static void addContextEdges(PathPieces &pieces, const LocationContext *LC) { 1420 const ParentMap &PM = LC->getParentMap(); 1421 PathPieces::iterator Prev = pieces.end(); 1422 for (PathPieces::iterator I = pieces.begin(), E = Prev; I != E; 1423 Prev = I, ++I) { 1424 auto *Piece = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1425 1426 if (!Piece) 1427 continue; 1428 1429 PathDiagnosticLocation SrcLoc = Piece->getStartLocation(); 1430 SmallVector<PathDiagnosticLocation, 4> SrcContexts; 1431 1432 PathDiagnosticLocation NextSrcContext = SrcLoc; 1433 const Stmt *InnerStmt = nullptr; 1434 while (NextSrcContext.isValid() && NextSrcContext.asStmt() != InnerStmt) { 1435 SrcContexts.push_back(NextSrcContext); 1436 InnerStmt = NextSrcContext.asStmt(); 1437 NextSrcContext = getEnclosingStmtLocation(InnerStmt, LC, 1438 /*allowNested=*/true); 1439 } 1440 1441 // Repeatedly split the edge as necessary. 1442 // This is important for nested logical expressions (||, &&, ?:) where we 1443 // want to show all the levels of context. 1444 while (true) { 1445 const Stmt *Dst = Piece->getEndLocation().getStmtOrNull(); 1446 1447 // We are looking at an edge. Is the destination within a larger 1448 // expression? 1449 PathDiagnosticLocation DstContext = 1450 getEnclosingStmtLocation(Dst, LC, /*allowNested=*/true); 1451 if (!DstContext.isValid() || DstContext.asStmt() == Dst) 1452 break; 1453 1454 // If the source is in the same context, we're already good. 1455 if (llvm::is_contained(SrcContexts, DstContext)) 1456 break; 1457 1458 // Update the subexpression node to point to the context edge. 1459 Piece->setStartLocation(DstContext); 1460 1461 // Try to extend the previous edge if it's at the same level as the source 1462 // context. 1463 if (Prev != E) { 1464 auto *PrevPiece = dyn_cast<PathDiagnosticControlFlowPiece>(Prev->get()); 1465 1466 if (PrevPiece) { 1467 if (const Stmt *PrevSrc = 1468 PrevPiece->getStartLocation().getStmtOrNull()) { 1469 const Stmt *PrevSrcParent = getStmtParent(PrevSrc, PM); 1470 if (PrevSrcParent == 1471 getStmtParent(DstContext.getStmtOrNull(), PM)) { 1472 PrevPiece->setEndLocation(DstContext); 1473 break; 1474 } 1475 } 1476 } 1477 } 1478 1479 // Otherwise, split the current edge into a context edge and a 1480 // subexpression edge. Note that the context statement may itself have 1481 // context. 1482 auto P = 1483 std::make_shared<PathDiagnosticControlFlowPiece>(SrcLoc, DstContext); 1484 Piece = P.get(); 1485 I = pieces.insert(I, std::move(P)); 1486 } 1487 } 1488 } 1489 1490 /// Move edges from a branch condition to a branch target 1491 /// when the condition is simple. 1492 /// 1493 /// This restructures some of the work of addContextEdges. That function 1494 /// creates edges this may destroy, but they work together to create a more 1495 /// aesthetically set of edges around branches. After the call to 1496 /// addContextEdges, we may have (1) an edge to the branch, (2) an edge from 1497 /// the branch to the branch condition, and (3) an edge from the branch 1498 /// condition to the branch target. We keep (1), but may wish to remove (2) 1499 /// and move the source of (3) to the branch if the branch condition is simple. 1500 static void simplifySimpleBranches(PathPieces &pieces) { 1501 for (PathPieces::iterator I = pieces.begin(), E = pieces.end(); I != E; ++I) { 1502 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1503 1504 if (!PieceI) 1505 continue; 1506 1507 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1508 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1509 1510 if (!s1Start || !s1End) 1511 continue; 1512 1513 PathPieces::iterator NextI = I; ++NextI; 1514 if (NextI == E) 1515 break; 1516 1517 PathDiagnosticControlFlowPiece *PieceNextI = nullptr; 1518 1519 while (true) { 1520 if (NextI == E) 1521 break; 1522 1523 const auto *EV = dyn_cast<PathDiagnosticEventPiece>(NextI->get()); 1524 if (EV) { 1525 StringRef S = EV->getString(); 1526 if (S == StrEnteringLoop || S == StrLoopBodyZero || 1527 S == StrLoopCollectionEmpty || S == StrLoopRangeEmpty) { 1528 ++NextI; 1529 continue; 1530 } 1531 break; 1532 } 1533 1534 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1535 break; 1536 } 1537 1538 if (!PieceNextI) 1539 continue; 1540 1541 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1542 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1543 1544 if (!s2Start || !s2End || s1End != s2Start) 1545 continue; 1546 1547 // We only perform this transformation for specific branch kinds. 1548 // We don't want to do this for do..while, for example. 1549 if (!isa<ForStmt, WhileStmt, IfStmt, ObjCForCollectionStmt, 1550 CXXForRangeStmt>(s1Start)) 1551 continue; 1552 1553 // Is s1End the branch condition? 1554 if (!isConditionForTerminator(s1Start, s1End)) 1555 continue; 1556 1557 // Perform the hoisting by eliminating (2) and changing the start 1558 // location of (3). 1559 PieceNextI->setStartLocation(PieceI->getStartLocation()); 1560 I = pieces.erase(I); 1561 } 1562 } 1563 1564 /// Returns the number of bytes in the given (character-based) SourceRange. 1565 /// 1566 /// If the locations in the range are not on the same line, returns 1567 /// std::nullopt. 1568 /// 1569 /// Note that this does not do a precise user-visible character or column count. 1570 static std::optional<size_t> getLengthOnSingleLine(const SourceManager &SM, 1571 SourceRange Range) { 1572 SourceRange ExpansionRange(SM.getExpansionLoc(Range.getBegin()), 1573 SM.getExpansionRange(Range.getEnd()).getEnd()); 1574 1575 FileID FID = SM.getFileID(ExpansionRange.getBegin()); 1576 if (FID != SM.getFileID(ExpansionRange.getEnd())) 1577 return std::nullopt; 1578 1579 std::optional<MemoryBufferRef> Buffer = SM.getBufferOrNone(FID); 1580 if (!Buffer) 1581 return std::nullopt; 1582 1583 unsigned BeginOffset = SM.getFileOffset(ExpansionRange.getBegin()); 1584 unsigned EndOffset = SM.getFileOffset(ExpansionRange.getEnd()); 1585 StringRef Snippet = Buffer->getBuffer().slice(BeginOffset, EndOffset); 1586 1587 // We're searching the raw bytes of the buffer here, which might include 1588 // escaped newlines and such. That's okay; we're trying to decide whether the 1589 // SourceRange is covering a large or small amount of space in the user's 1590 // editor. 1591 if (Snippet.find_first_of("\r\n") != StringRef::npos) 1592 return std::nullopt; 1593 1594 // This isn't Unicode-aware, but it doesn't need to be. 1595 return Snippet.size(); 1596 } 1597 1598 /// \sa getLengthOnSingleLine(SourceManager, SourceRange) 1599 static std::optional<size_t> getLengthOnSingleLine(const SourceManager &SM, 1600 const Stmt *S) { 1601 return getLengthOnSingleLine(SM, S->getSourceRange()); 1602 } 1603 1604 /// Eliminate two-edge cycles created by addContextEdges(). 1605 /// 1606 /// Once all the context edges are in place, there are plenty of cases where 1607 /// there's a single edge from a top-level statement to a subexpression, 1608 /// followed by a single path note, and then a reverse edge to get back out to 1609 /// the top level. If the statement is simple enough, the subexpression edges 1610 /// just add noise and make it harder to understand what's going on. 1611 /// 1612 /// This function only removes edges in pairs, because removing only one edge 1613 /// might leave other edges dangling. 1614 /// 1615 /// This will not remove edges in more complicated situations: 1616 /// - if there is more than one "hop" leading to or from a subexpression. 1617 /// - if there is an inlined call between the edges instead of a single event. 1618 /// - if the whole statement is large enough that having subexpression arrows 1619 /// might be helpful. 1620 static void removeContextCycles(PathPieces &Path, const SourceManager &SM) { 1621 for (PathPieces::iterator I = Path.begin(), E = Path.end(); I != E; ) { 1622 // Pattern match the current piece and its successor. 1623 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1624 1625 if (!PieceI) { 1626 ++I; 1627 continue; 1628 } 1629 1630 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1631 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1632 1633 PathPieces::iterator NextI = I; ++NextI; 1634 if (NextI == E) 1635 break; 1636 1637 const auto *PieceNextI = 1638 dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1639 1640 if (!PieceNextI) { 1641 if (isa<PathDiagnosticEventPiece>(NextI->get())) { 1642 ++NextI; 1643 if (NextI == E) 1644 break; 1645 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1646 } 1647 1648 if (!PieceNextI) { 1649 ++I; 1650 continue; 1651 } 1652 } 1653 1654 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1655 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1656 1657 if (s1Start && s2Start && s1Start == s2End && s2Start == s1End) { 1658 const size_t MAX_SHORT_LINE_LENGTH = 80; 1659 std::optional<size_t> s1Length = getLengthOnSingleLine(SM, s1Start); 1660 if (s1Length && *s1Length <= MAX_SHORT_LINE_LENGTH) { 1661 std::optional<size_t> s2Length = getLengthOnSingleLine(SM, s2Start); 1662 if (s2Length && *s2Length <= MAX_SHORT_LINE_LENGTH) { 1663 Path.erase(I); 1664 I = Path.erase(NextI); 1665 continue; 1666 } 1667 } 1668 } 1669 1670 ++I; 1671 } 1672 } 1673 1674 /// Return true if X is contained by Y. 1675 static bool lexicalContains(const ParentMap &PM, const Stmt *X, const Stmt *Y) { 1676 while (X) { 1677 if (X == Y) 1678 return true; 1679 X = PM.getParent(X); 1680 } 1681 return false; 1682 } 1683 1684 // Remove short edges on the same line less than 3 columns in difference. 1685 static void removePunyEdges(PathPieces &path, const SourceManager &SM, 1686 const ParentMap &PM) { 1687 bool erased = false; 1688 1689 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; 1690 erased ? I : ++I) { 1691 erased = false; 1692 1693 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1694 1695 if (!PieceI) 1696 continue; 1697 1698 const Stmt *start = PieceI->getStartLocation().getStmtOrNull(); 1699 const Stmt *end = PieceI->getEndLocation().getStmtOrNull(); 1700 1701 if (!start || !end) 1702 continue; 1703 1704 const Stmt *endParent = PM.getParent(end); 1705 if (!endParent) 1706 continue; 1707 1708 if (isConditionForTerminator(end, endParent)) 1709 continue; 1710 1711 SourceLocation FirstLoc = start->getBeginLoc(); 1712 SourceLocation SecondLoc = end->getBeginLoc(); 1713 1714 if (!SM.isWrittenInSameFile(FirstLoc, SecondLoc)) 1715 continue; 1716 if (SM.isBeforeInTranslationUnit(SecondLoc, FirstLoc)) 1717 std::swap(SecondLoc, FirstLoc); 1718 1719 SourceRange EdgeRange(FirstLoc, SecondLoc); 1720 std::optional<size_t> ByteWidth = getLengthOnSingleLine(SM, EdgeRange); 1721 1722 // If the statements are on different lines, continue. 1723 if (!ByteWidth) 1724 continue; 1725 1726 const size_t MAX_PUNY_EDGE_LENGTH = 2; 1727 if (*ByteWidth <= MAX_PUNY_EDGE_LENGTH) { 1728 // FIXME: There are enough /bytes/ between the endpoints of the edge, but 1729 // there might not be enough /columns/. A proper user-visible column count 1730 // is probably too expensive, though. 1731 I = path.erase(I); 1732 erased = true; 1733 continue; 1734 } 1735 } 1736 } 1737 1738 static void removeIdenticalEvents(PathPieces &path) { 1739 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ++I) { 1740 const auto *PieceI = dyn_cast<PathDiagnosticEventPiece>(I->get()); 1741 1742 if (!PieceI) 1743 continue; 1744 1745 PathPieces::iterator NextI = I; ++NextI; 1746 if (NextI == E) 1747 return; 1748 1749 const auto *PieceNextI = dyn_cast<PathDiagnosticEventPiece>(NextI->get()); 1750 1751 if (!PieceNextI) 1752 continue; 1753 1754 // Erase the second piece if it has the same exact message text. 1755 if (PieceI->getString() == PieceNextI->getString()) { 1756 path.erase(NextI); 1757 } 1758 } 1759 } 1760 1761 static bool optimizeEdges(const PathDiagnosticConstruct &C, PathPieces &path, 1762 OptimizedCallsSet &OCS) { 1763 bool hasChanges = false; 1764 const LocationContext *LC = C.getLocationContextFor(&path); 1765 assert(LC); 1766 const ParentMap &PM = LC->getParentMap(); 1767 const SourceManager &SM = C.getSourceManager(); 1768 1769 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ) { 1770 // Optimize subpaths. 1771 if (auto *CallI = dyn_cast<PathDiagnosticCallPiece>(I->get())) { 1772 // Record the fact that a call has been optimized so we only do the 1773 // effort once. 1774 if (!OCS.count(CallI)) { 1775 while (optimizeEdges(C, CallI->path, OCS)) { 1776 } 1777 OCS.insert(CallI); 1778 } 1779 ++I; 1780 continue; 1781 } 1782 1783 // Pattern match the current piece and its successor. 1784 auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1785 1786 if (!PieceI) { 1787 ++I; 1788 continue; 1789 } 1790 1791 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1792 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1793 const Stmt *level1 = getStmtParent(s1Start, PM); 1794 const Stmt *level2 = getStmtParent(s1End, PM); 1795 1796 PathPieces::iterator NextI = I; ++NextI; 1797 if (NextI == E) 1798 break; 1799 1800 const auto *PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1801 1802 if (!PieceNextI) { 1803 ++I; 1804 continue; 1805 } 1806 1807 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1808 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1809 const Stmt *level3 = getStmtParent(s2Start, PM); 1810 const Stmt *level4 = getStmtParent(s2End, PM); 1811 1812 // Rule I. 1813 // 1814 // If we have two consecutive control edges whose end/begin locations 1815 // are at the same level (e.g. statements or top-level expressions within 1816 // a compound statement, or siblings share a single ancestor expression), 1817 // then merge them if they have no interesting intermediate event. 1818 // 1819 // For example: 1820 // 1821 // (1.1 -> 1.2) -> (1.2 -> 1.3) becomes (1.1 -> 1.3) because the common 1822 // parent is '1'. Here 'x.y.z' represents the hierarchy of statements. 1823 // 1824 // NOTE: this will be limited later in cases where we add barriers 1825 // to prevent this optimization. 1826 if (level1 && level1 == level2 && level1 == level3 && level1 == level4) { 1827 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1828 path.erase(NextI); 1829 hasChanges = true; 1830 continue; 1831 } 1832 1833 // Rule II. 1834 // 1835 // Eliminate edges between subexpressions and parent expressions 1836 // when the subexpression is consumed. 1837 // 1838 // NOTE: this will be limited later in cases where we add barriers 1839 // to prevent this optimization. 1840 if (s1End && s1End == s2Start && level2) { 1841 bool removeEdge = false; 1842 // Remove edges into the increment or initialization of a 1843 // loop that have no interleaving event. This means that 1844 // they aren't interesting. 1845 if (isIncrementOrInitInForLoop(s1End, level2)) 1846 removeEdge = true; 1847 // Next only consider edges that are not anchored on 1848 // the condition of a terminator. This are intermediate edges 1849 // that we might want to trim. 1850 else if (!isConditionForTerminator(level2, s1End)) { 1851 // Trim edges on expressions that are consumed by 1852 // the parent expression. 1853 if (isa<Expr>(s1End) && PM.isConsumedExpr(cast<Expr>(s1End))) { 1854 removeEdge = true; 1855 } 1856 // Trim edges where a lexical containment doesn't exist. 1857 // For example: 1858 // 1859 // X -> Y -> Z 1860 // 1861 // If 'Z' lexically contains Y (it is an ancestor) and 1862 // 'X' does not lexically contain Y (it is a descendant OR 1863 // it has no lexical relationship at all) then trim. 1864 // 1865 // This can eliminate edges where we dive into a subexpression 1866 // and then pop back out, etc. 1867 else if (s1Start && s2End && 1868 lexicalContains(PM, s2Start, s2End) && 1869 !lexicalContains(PM, s1End, s1Start)) { 1870 removeEdge = true; 1871 } 1872 // Trim edges from a subexpression back to the top level if the 1873 // subexpression is on a different line. 1874 // 1875 // A.1 -> A -> B 1876 // becomes 1877 // A.1 -> B 1878 // 1879 // These edges just look ugly and don't usually add anything. 1880 else if (s1Start && s2End && 1881 lexicalContains(PM, s1Start, s1End)) { 1882 SourceRange EdgeRange(PieceI->getEndLocation().asLocation(), 1883 PieceI->getStartLocation().asLocation()); 1884 if (!getLengthOnSingleLine(SM, EdgeRange)) 1885 removeEdge = true; 1886 } 1887 } 1888 1889 if (removeEdge) { 1890 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1891 path.erase(NextI); 1892 hasChanges = true; 1893 continue; 1894 } 1895 } 1896 1897 // Optimize edges for ObjC fast-enumeration loops. 1898 // 1899 // (X -> collection) -> (collection -> element) 1900 // 1901 // becomes: 1902 // 1903 // (X -> element) 1904 if (s1End == s2Start) { 1905 const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(level3); 1906 if (FS && FS->getCollection()->IgnoreParens() == s2Start && 1907 s2End == FS->getElement()) { 1908 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1909 path.erase(NextI); 1910 hasChanges = true; 1911 continue; 1912 } 1913 } 1914 1915 // No changes at this index? Move to the next one. 1916 ++I; 1917 } 1918 1919 if (!hasChanges) { 1920 // Adjust edges into subexpressions to make them more uniform 1921 // and aesthetically pleasing. 1922 addContextEdges(path, LC); 1923 // Remove "cyclical" edges that include one or more context edges. 1924 removeContextCycles(path, SM); 1925 // Hoist edges originating from branch conditions to branches 1926 // for simple branches. 1927 simplifySimpleBranches(path); 1928 // Remove any puny edges left over after primary optimization pass. 1929 removePunyEdges(path, SM, PM); 1930 // Remove identical events. 1931 removeIdenticalEvents(path); 1932 } 1933 1934 return hasChanges; 1935 } 1936 1937 /// Drop the very first edge in a path, which should be a function entry edge. 1938 /// 1939 /// If the first edge is not a function entry edge (say, because the first 1940 /// statement had an invalid source location), this function does nothing. 1941 // FIXME: We should just generate invalid edges anyway and have the optimizer 1942 // deal with them. 1943 static void dropFunctionEntryEdge(const PathDiagnosticConstruct &C, 1944 PathPieces &Path) { 1945 const auto *FirstEdge = 1946 dyn_cast<PathDiagnosticControlFlowPiece>(Path.front().get()); 1947 if (!FirstEdge) 1948 return; 1949 1950 const Decl *D = C.getLocationContextFor(&Path)->getDecl(); 1951 PathDiagnosticLocation EntryLoc = 1952 PathDiagnosticLocation::createBegin(D, C.getSourceManager()); 1953 if (FirstEdge->getStartLocation() != EntryLoc) 1954 return; 1955 1956 Path.pop_front(); 1957 } 1958 1959 /// Populate executes lines with lines containing at least one diagnostics. 1960 static void updateExecutedLinesWithDiagnosticPieces(PathDiagnostic &PD) { 1961 1962 PathPieces path = PD.path.flatten(/*ShouldFlattenMacros=*/true); 1963 FilesToLineNumsMap &ExecutedLines = PD.getExecutedLines(); 1964 1965 for (const auto &P : path) { 1966 FullSourceLoc Loc = P->getLocation().asLocation().getExpansionLoc(); 1967 FileID FID = Loc.getFileID(); 1968 unsigned LineNo = Loc.getLineNumber(); 1969 assert(FID.isValid()); 1970 ExecutedLines[FID].insert(LineNo); 1971 } 1972 } 1973 1974 PathDiagnosticConstruct::PathDiagnosticConstruct( 1975 const PathDiagnosticConsumer *PDC, const ExplodedNode *ErrorNode, 1976 const PathSensitiveBugReport *R) 1977 : Consumer(PDC), CurrentNode(ErrorNode), 1978 SM(CurrentNode->getCodeDecl().getASTContext().getSourceManager()), 1979 PD(generateEmptyDiagnosticForReport(R, getSourceManager())) { 1980 LCM[&PD->getActivePath()] = ErrorNode->getLocationContext(); 1981 } 1982 1983 PathDiagnosticBuilder::PathDiagnosticBuilder( 1984 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath, 1985 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode, 1986 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics) 1987 : BugReporterContext(BRC), BugPath(std::move(BugPath)), R(r), 1988 ErrorNode(ErrorNode), 1989 VisitorsDiagnostics(std::move(VisitorsDiagnostics)) {} 1990 1991 std::unique_ptr<PathDiagnostic> 1992 PathDiagnosticBuilder::generate(const PathDiagnosticConsumer *PDC) const { 1993 PathDiagnosticConstruct Construct(PDC, ErrorNode, R); 1994 1995 const SourceManager &SM = getSourceManager(); 1996 const AnalyzerOptions &Opts = getAnalyzerOptions(); 1997 1998 if (!PDC->shouldGenerateDiagnostics()) 1999 return generateEmptyDiagnosticForReport(R, getSourceManager()); 2000 2001 // Construct the final (warning) event for the bug report. 2002 auto EndNotes = VisitorsDiagnostics->find(ErrorNode); 2003 PathDiagnosticPieceRef LastPiece; 2004 if (EndNotes != VisitorsDiagnostics->end()) { 2005 assert(!EndNotes->second.empty()); 2006 LastPiece = EndNotes->second[0]; 2007 } else { 2008 LastPiece = BugReporterVisitor::getDefaultEndPath(*this, ErrorNode, 2009 *getBugReport()); 2010 } 2011 Construct.PD->setEndOfPath(LastPiece); 2012 2013 PathDiagnosticLocation PrevLoc = Construct.PD->getLocation(); 2014 // From the error node to the root, ascend the bug path and construct the bug 2015 // report. 2016 while (Construct.ascendToPrevNode()) { 2017 generatePathDiagnosticsForNode(Construct, PrevLoc); 2018 2019 auto VisitorNotes = VisitorsDiagnostics->find(Construct.getCurrentNode()); 2020 if (VisitorNotes == VisitorsDiagnostics->end()) 2021 continue; 2022 2023 // This is a workaround due to inability to put shared PathDiagnosticPiece 2024 // into a FoldingSet. 2025 std::set<llvm::FoldingSetNodeID> DeduplicationSet; 2026 2027 // Add pieces from custom visitors. 2028 for (const PathDiagnosticPieceRef &Note : VisitorNotes->second) { 2029 llvm::FoldingSetNodeID ID; 2030 Note->Profile(ID); 2031 if (!DeduplicationSet.insert(ID).second) 2032 continue; 2033 2034 if (PDC->shouldAddPathEdges()) 2035 addEdgeToPath(Construct.getActivePath(), PrevLoc, Note->getLocation()); 2036 updateStackPiecesWithMessage(Note, Construct.CallStack); 2037 Construct.getActivePath().push_front(Note); 2038 } 2039 } 2040 2041 if (PDC->shouldAddPathEdges()) { 2042 // Add an edge to the start of the function. 2043 // We'll prune it out later, but it helps make diagnostics more uniform. 2044 const StackFrameContext *CalleeLC = 2045 Construct.getLocationContextForActivePath()->getStackFrame(); 2046 const Decl *D = CalleeLC->getDecl(); 2047 addEdgeToPath(Construct.getActivePath(), PrevLoc, 2048 PathDiagnosticLocation::createBegin(D, SM)); 2049 } 2050 2051 2052 // Finally, prune the diagnostic path of uninteresting stuff. 2053 if (!Construct.PD->path.empty()) { 2054 if (R->shouldPrunePath() && Opts.ShouldPrunePaths) { 2055 bool stillHasNotes = 2056 removeUnneededCalls(Construct, Construct.getMutablePieces(), R); 2057 assert(stillHasNotes); 2058 (void)stillHasNotes; 2059 } 2060 2061 // Remove pop-up notes if needed. 2062 if (!Opts.ShouldAddPopUpNotes) 2063 removePopUpNotes(Construct.getMutablePieces()); 2064 2065 // Redirect all call pieces to have valid locations. 2066 adjustCallLocations(Construct.getMutablePieces()); 2067 removePiecesWithInvalidLocations(Construct.getMutablePieces()); 2068 2069 if (PDC->shouldAddPathEdges()) { 2070 2071 // Reduce the number of edges from a very conservative set 2072 // to an aesthetically pleasing subset that conveys the 2073 // necessary information. 2074 OptimizedCallsSet OCS; 2075 while (optimizeEdges(Construct, Construct.getMutablePieces(), OCS)) { 2076 } 2077 2078 // Drop the very first function-entry edge. It's not really necessary 2079 // for top-level functions. 2080 dropFunctionEntryEdge(Construct, Construct.getMutablePieces()); 2081 } 2082 2083 // Remove messages that are basically the same, and edges that may not 2084 // make sense. 2085 // We have to do this after edge optimization in the Extensive mode. 2086 removeRedundantMsgs(Construct.getMutablePieces()); 2087 removeEdgesToDefaultInitializers(Construct.getMutablePieces()); 2088 } 2089 2090 if (Opts.ShouldDisplayMacroExpansions) 2091 CompactMacroExpandedPieces(Construct.getMutablePieces(), SM); 2092 2093 return std::move(Construct.PD); 2094 } 2095 2096 //===----------------------------------------------------------------------===// 2097 // Methods for BugType and subclasses. 2098 //===----------------------------------------------------------------------===// 2099 2100 void BugType::anchor() {} 2101 2102 void BuiltinBug::anchor() {} 2103 2104 //===----------------------------------------------------------------------===// 2105 // Methods for BugReport and subclasses. 2106 //===----------------------------------------------------------------------===// 2107 2108 LLVM_ATTRIBUTE_USED static bool 2109 isDependency(const CheckerRegistryData &Registry, StringRef CheckerName) { 2110 for (const std::pair<StringRef, StringRef> &Pair : Registry.Dependencies) { 2111 if (Pair.second == CheckerName) 2112 return true; 2113 } 2114 return false; 2115 } 2116 2117 LLVM_ATTRIBUTE_USED static bool isHidden(const CheckerRegistryData &Registry, 2118 StringRef CheckerName) { 2119 for (const CheckerInfo &Checker : Registry.Checkers) { 2120 if (Checker.FullName == CheckerName) 2121 return Checker.IsHidden; 2122 } 2123 llvm_unreachable( 2124 "Checker name not found in CheckerRegistry -- did you retrieve it " 2125 "correctly from CheckerManager::getCurrentCheckerName?"); 2126 } 2127 2128 PathSensitiveBugReport::PathSensitiveBugReport( 2129 const BugType &bt, StringRef shortDesc, StringRef desc, 2130 const ExplodedNode *errorNode, PathDiagnosticLocation LocationToUnique, 2131 const Decl *DeclToUnique) 2132 : BugReport(Kind::PathSensitive, bt, shortDesc, desc), ErrorNode(errorNode), 2133 ErrorNodeRange(getStmt() ? getStmt()->getSourceRange() : SourceRange()), 2134 UniqueingLocation(LocationToUnique), UniqueingDecl(DeclToUnique) { 2135 assert(!isDependency(ErrorNode->getState() 2136 ->getAnalysisManager() 2137 .getCheckerManager() 2138 ->getCheckerRegistryData(), 2139 bt.getCheckerName()) && 2140 "Some checkers depend on this one! We don't allow dependency " 2141 "checkers to emit warnings, because checkers should depend on " 2142 "*modeling*, not *diagnostics*."); 2143 2144 assert( 2145 (bt.getCheckerName().startswith("debug") || 2146 !isHidden(ErrorNode->getState() 2147 ->getAnalysisManager() 2148 .getCheckerManager() 2149 ->getCheckerRegistryData(), 2150 bt.getCheckerName())) && 2151 "Hidden checkers musn't emit diagnostics as they are by definition " 2152 "non-user facing!"); 2153 } 2154 2155 void PathSensitiveBugReport::addVisitor( 2156 std::unique_ptr<BugReporterVisitor> visitor) { 2157 if (!visitor) 2158 return; 2159 2160 llvm::FoldingSetNodeID ID; 2161 visitor->Profile(ID); 2162 2163 void *InsertPos = nullptr; 2164 if (CallbacksSet.FindNodeOrInsertPos(ID, InsertPos)) { 2165 return; 2166 } 2167 2168 Callbacks.push_back(std::move(visitor)); 2169 } 2170 2171 void PathSensitiveBugReport::clearVisitors() { 2172 Callbacks.clear(); 2173 } 2174 2175 const Decl *PathSensitiveBugReport::getDeclWithIssue() const { 2176 const ExplodedNode *N = getErrorNode(); 2177 if (!N) 2178 return nullptr; 2179 2180 const LocationContext *LC = N->getLocationContext(); 2181 return LC->getStackFrame()->getDecl(); 2182 } 2183 2184 void BasicBugReport::Profile(llvm::FoldingSetNodeID& hash) const { 2185 hash.AddInteger(static_cast<int>(getKind())); 2186 hash.AddPointer(&BT); 2187 hash.AddString(Description); 2188 assert(Location.isValid()); 2189 Location.Profile(hash); 2190 2191 for (SourceRange range : Ranges) { 2192 if (!range.isValid()) 2193 continue; 2194 hash.Add(range.getBegin()); 2195 hash.Add(range.getEnd()); 2196 } 2197 } 2198 2199 void PathSensitiveBugReport::Profile(llvm::FoldingSetNodeID &hash) const { 2200 hash.AddInteger(static_cast<int>(getKind())); 2201 hash.AddPointer(&BT); 2202 hash.AddString(Description); 2203 PathDiagnosticLocation UL = getUniqueingLocation(); 2204 if (UL.isValid()) { 2205 UL.Profile(hash); 2206 } else { 2207 // TODO: The statement may be null if the report was emitted before any 2208 // statements were executed. In particular, some checkers by design 2209 // occasionally emit their reports in empty functions (that have no 2210 // statements in their body). Do we profile correctly in this case? 2211 hash.AddPointer(ErrorNode->getCurrentOrPreviousStmtForDiagnostics()); 2212 } 2213 2214 for (SourceRange range : Ranges) { 2215 if (!range.isValid()) 2216 continue; 2217 hash.Add(range.getBegin()); 2218 hash.Add(range.getEnd()); 2219 } 2220 } 2221 2222 template <class T> 2223 static void insertToInterestingnessMap( 2224 llvm::DenseMap<T, bugreporter::TrackingKind> &InterestingnessMap, T Val, 2225 bugreporter::TrackingKind TKind) { 2226 auto Result = InterestingnessMap.insert({Val, TKind}); 2227 2228 if (Result.second) 2229 return; 2230 2231 // Even if this symbol/region was already marked as interesting as a 2232 // condition, if we later mark it as interesting again but with 2233 // thorough tracking, overwrite it. Entities marked with thorough 2234 // interestiness are the most important (or most interesting, if you will), 2235 // and we wouldn't like to downplay their importance. 2236 2237 switch (TKind) { 2238 case bugreporter::TrackingKind::Thorough: 2239 Result.first->getSecond() = bugreporter::TrackingKind::Thorough; 2240 return; 2241 case bugreporter::TrackingKind::Condition: 2242 return; 2243 } 2244 2245 llvm_unreachable( 2246 "BugReport::markInteresting currently can only handle 2 different " 2247 "tracking kinds! Please define what tracking kind should this entitiy" 2248 "have, if it was already marked as interesting with a different kind!"); 2249 } 2250 2251 void PathSensitiveBugReport::markInteresting(SymbolRef sym, 2252 bugreporter::TrackingKind TKind) { 2253 if (!sym) 2254 return; 2255 2256 insertToInterestingnessMap(InterestingSymbols, sym, TKind); 2257 2258 // FIXME: No tests exist for this code and it is questionable: 2259 // How to handle multiple metadata for the same region? 2260 if (const auto *meta = dyn_cast<SymbolMetadata>(sym)) 2261 markInteresting(meta->getRegion(), TKind); 2262 } 2263 2264 void PathSensitiveBugReport::markNotInteresting(SymbolRef sym) { 2265 if (!sym) 2266 return; 2267 InterestingSymbols.erase(sym); 2268 2269 // The metadata part of markInteresting is not reversed here. 2270 // Just making the same region not interesting is incorrect 2271 // in specific cases. 2272 if (const auto *meta = dyn_cast<SymbolMetadata>(sym)) 2273 markNotInteresting(meta->getRegion()); 2274 } 2275 2276 void PathSensitiveBugReport::markInteresting(const MemRegion *R, 2277 bugreporter::TrackingKind TKind) { 2278 if (!R) 2279 return; 2280 2281 R = R->getBaseRegion(); 2282 insertToInterestingnessMap(InterestingRegions, R, TKind); 2283 2284 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2285 markInteresting(SR->getSymbol(), TKind); 2286 } 2287 2288 void PathSensitiveBugReport::markNotInteresting(const MemRegion *R) { 2289 if (!R) 2290 return; 2291 2292 R = R->getBaseRegion(); 2293 InterestingRegions.erase(R); 2294 2295 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2296 markNotInteresting(SR->getSymbol()); 2297 } 2298 2299 void PathSensitiveBugReport::markInteresting(SVal V, 2300 bugreporter::TrackingKind TKind) { 2301 markInteresting(V.getAsRegion(), TKind); 2302 markInteresting(V.getAsSymbol(), TKind); 2303 } 2304 2305 void PathSensitiveBugReport::markInteresting(const LocationContext *LC) { 2306 if (!LC) 2307 return; 2308 InterestingLocationContexts.insert(LC); 2309 } 2310 2311 std::optional<bugreporter::TrackingKind> 2312 PathSensitiveBugReport::getInterestingnessKind(SVal V) const { 2313 auto RKind = getInterestingnessKind(V.getAsRegion()); 2314 auto SKind = getInterestingnessKind(V.getAsSymbol()); 2315 if (!RKind) 2316 return SKind; 2317 if (!SKind) 2318 return RKind; 2319 2320 // If either is marked with throrough tracking, return that, we wouldn't like 2321 // to downplay a note's importance by 'only' mentioning it as a condition. 2322 switch(*RKind) { 2323 case bugreporter::TrackingKind::Thorough: 2324 return RKind; 2325 case bugreporter::TrackingKind::Condition: 2326 return SKind; 2327 } 2328 2329 llvm_unreachable( 2330 "BugReport::getInterestingnessKind currently can only handle 2 different " 2331 "tracking kinds! Please define what tracking kind should we return here " 2332 "when the kind of getAsRegion() and getAsSymbol() is different!"); 2333 return std::nullopt; 2334 } 2335 2336 std::optional<bugreporter::TrackingKind> 2337 PathSensitiveBugReport::getInterestingnessKind(SymbolRef sym) const { 2338 if (!sym) 2339 return std::nullopt; 2340 // We don't currently consider metadata symbols to be interesting 2341 // even if we know their region is interesting. Is that correct behavior? 2342 auto It = InterestingSymbols.find(sym); 2343 if (It == InterestingSymbols.end()) 2344 return std::nullopt; 2345 return It->getSecond(); 2346 } 2347 2348 std::optional<bugreporter::TrackingKind> 2349 PathSensitiveBugReport::getInterestingnessKind(const MemRegion *R) const { 2350 if (!R) 2351 return std::nullopt; 2352 2353 R = R->getBaseRegion(); 2354 auto It = InterestingRegions.find(R); 2355 if (It != InterestingRegions.end()) 2356 return It->getSecond(); 2357 2358 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2359 return getInterestingnessKind(SR->getSymbol()); 2360 return std::nullopt; 2361 } 2362 2363 bool PathSensitiveBugReport::isInteresting(SVal V) const { 2364 return getInterestingnessKind(V).has_value(); 2365 } 2366 2367 bool PathSensitiveBugReport::isInteresting(SymbolRef sym) const { 2368 return getInterestingnessKind(sym).has_value(); 2369 } 2370 2371 bool PathSensitiveBugReport::isInteresting(const MemRegion *R) const { 2372 return getInterestingnessKind(R).has_value(); 2373 } 2374 2375 bool PathSensitiveBugReport::isInteresting(const LocationContext *LC) const { 2376 if (!LC) 2377 return false; 2378 return InterestingLocationContexts.count(LC); 2379 } 2380 2381 const Stmt *PathSensitiveBugReport::getStmt() const { 2382 if (!ErrorNode) 2383 return nullptr; 2384 2385 ProgramPoint ProgP = ErrorNode->getLocation(); 2386 const Stmt *S = nullptr; 2387 2388 if (std::optional<BlockEntrance> BE = ProgP.getAs<BlockEntrance>()) { 2389 CFGBlock &Exit = ProgP.getLocationContext()->getCFG()->getExit(); 2390 if (BE->getBlock() == &Exit) 2391 S = ErrorNode->getPreviousStmtForDiagnostics(); 2392 } 2393 if (!S) 2394 S = ErrorNode->getStmtForDiagnostics(); 2395 2396 return S; 2397 } 2398 2399 ArrayRef<SourceRange> 2400 PathSensitiveBugReport::getRanges() const { 2401 // If no custom ranges, add the range of the statement corresponding to 2402 // the error node. 2403 if (Ranges.empty() && isa_and_nonnull<Expr>(getStmt())) 2404 return ErrorNodeRange; 2405 2406 return Ranges; 2407 } 2408 2409 PathDiagnosticLocation 2410 PathSensitiveBugReport::getLocation() const { 2411 assert(ErrorNode && "Cannot create a location with a null node."); 2412 const Stmt *S = ErrorNode->getStmtForDiagnostics(); 2413 ProgramPoint P = ErrorNode->getLocation(); 2414 const LocationContext *LC = P.getLocationContext(); 2415 SourceManager &SM = 2416 ErrorNode->getState()->getStateManager().getContext().getSourceManager(); 2417 2418 if (!S) { 2419 // If this is an implicit call, return the implicit call point location. 2420 if (std::optional<PreImplicitCall> PIE = P.getAs<PreImplicitCall>()) 2421 return PathDiagnosticLocation(PIE->getLocation(), SM); 2422 if (auto FE = P.getAs<FunctionExitPoint>()) { 2423 if (const ReturnStmt *RS = FE->getStmt()) 2424 return PathDiagnosticLocation::createBegin(RS, SM, LC); 2425 } 2426 S = ErrorNode->getNextStmtForDiagnostics(); 2427 } 2428 2429 if (S) { 2430 // For member expressions, return the location of the '.' or '->'. 2431 if (const auto *ME = dyn_cast<MemberExpr>(S)) 2432 return PathDiagnosticLocation::createMemberLoc(ME, SM); 2433 2434 // For binary operators, return the location of the operator. 2435 if (const auto *B = dyn_cast<BinaryOperator>(S)) 2436 return PathDiagnosticLocation::createOperatorLoc(B, SM); 2437 2438 if (P.getAs<PostStmtPurgeDeadSymbols>()) 2439 return PathDiagnosticLocation::createEnd(S, SM, LC); 2440 2441 if (S->getBeginLoc().isValid()) 2442 return PathDiagnosticLocation(S, SM, LC); 2443 2444 return PathDiagnosticLocation( 2445 PathDiagnosticLocation::getValidSourceLocation(S, LC), SM); 2446 } 2447 2448 return PathDiagnosticLocation::createDeclEnd(ErrorNode->getLocationContext(), 2449 SM); 2450 } 2451 2452 //===----------------------------------------------------------------------===// 2453 // Methods for BugReporter and subclasses. 2454 //===----------------------------------------------------------------------===// 2455 2456 const ExplodedGraph &PathSensitiveBugReporter::getGraph() const { 2457 return Eng.getGraph(); 2458 } 2459 2460 ProgramStateManager &PathSensitiveBugReporter::getStateManager() const { 2461 return Eng.getStateManager(); 2462 } 2463 2464 BugReporter::BugReporter(BugReporterData &d) : D(d) {} 2465 BugReporter::~BugReporter() { 2466 // Make sure reports are flushed. 2467 assert(StrBugTypes.empty() && 2468 "Destroying BugReporter before diagnostics are emitted!"); 2469 2470 // Free the bug reports we are tracking. 2471 for (const auto I : EQClassesVector) 2472 delete I; 2473 } 2474 2475 void BugReporter::FlushReports() { 2476 // We need to flush reports in deterministic order to ensure the order 2477 // of the reports is consistent between runs. 2478 for (const auto EQ : EQClassesVector) 2479 FlushReport(*EQ); 2480 2481 // BugReporter owns and deletes only BugTypes created implicitly through 2482 // EmitBasicReport. 2483 // FIXME: There are leaks from checkers that assume that the BugTypes they 2484 // create will be destroyed by the BugReporter. 2485 StrBugTypes.clear(); 2486 } 2487 2488 //===----------------------------------------------------------------------===// 2489 // PathDiagnostics generation. 2490 //===----------------------------------------------------------------------===// 2491 2492 namespace { 2493 2494 /// A wrapper around an ExplodedGraph that contains a single path from the root 2495 /// to the error node. 2496 class BugPathInfo { 2497 public: 2498 std::unique_ptr<ExplodedGraph> BugPath; 2499 PathSensitiveBugReport *Report; 2500 const ExplodedNode *ErrorNode; 2501 }; 2502 2503 /// A wrapper around an ExplodedGraph whose leafs are all error nodes. Can 2504 /// conveniently retrieve bug paths from a single error node to the root. 2505 class BugPathGetter { 2506 std::unique_ptr<ExplodedGraph> TrimmedGraph; 2507 2508 using PriorityMapTy = llvm::DenseMap<const ExplodedNode *, unsigned>; 2509 2510 /// Assign each node with its distance from the root. 2511 PriorityMapTy PriorityMap; 2512 2513 /// Since the getErrorNode() or BugReport refers to the original ExplodedGraph, 2514 /// we need to pair it to the error node of the constructed trimmed graph. 2515 using ReportNewNodePair = 2516 std::pair<PathSensitiveBugReport *, const ExplodedNode *>; 2517 SmallVector<ReportNewNodePair, 32> ReportNodes; 2518 2519 BugPathInfo CurrentBugPath; 2520 2521 /// A helper class for sorting ExplodedNodes by priority. 2522 template <bool Descending> 2523 class PriorityCompare { 2524 const PriorityMapTy &PriorityMap; 2525 2526 public: 2527 PriorityCompare(const PriorityMapTy &M) : PriorityMap(M) {} 2528 2529 bool operator()(const ExplodedNode *LHS, const ExplodedNode *RHS) const { 2530 PriorityMapTy::const_iterator LI = PriorityMap.find(LHS); 2531 PriorityMapTy::const_iterator RI = PriorityMap.find(RHS); 2532 PriorityMapTy::const_iterator E = PriorityMap.end(); 2533 2534 if (LI == E) 2535 return Descending; 2536 if (RI == E) 2537 return !Descending; 2538 2539 return Descending ? LI->second > RI->second 2540 : LI->second < RI->second; 2541 } 2542 2543 bool operator()(const ReportNewNodePair &LHS, 2544 const ReportNewNodePair &RHS) const { 2545 return (*this)(LHS.second, RHS.second); 2546 } 2547 }; 2548 2549 public: 2550 BugPathGetter(const ExplodedGraph *OriginalGraph, 2551 ArrayRef<PathSensitiveBugReport *> &bugReports); 2552 2553 BugPathInfo *getNextBugPath(); 2554 }; 2555 2556 } // namespace 2557 2558 BugPathGetter::BugPathGetter(const ExplodedGraph *OriginalGraph, 2559 ArrayRef<PathSensitiveBugReport *> &bugReports) { 2560 SmallVector<const ExplodedNode *, 32> Nodes; 2561 for (const auto I : bugReports) { 2562 assert(I->isValid() && 2563 "We only allow BugReporterVisitors and BugReporter itself to " 2564 "invalidate reports!"); 2565 Nodes.emplace_back(I->getErrorNode()); 2566 } 2567 2568 // The trimmed graph is created in the body of the constructor to ensure 2569 // that the DenseMaps have been initialized already. 2570 InterExplodedGraphMap ForwardMap; 2571 TrimmedGraph = OriginalGraph->trim(Nodes, &ForwardMap); 2572 2573 // Find the (first) error node in the trimmed graph. We just need to consult 2574 // the node map which maps from nodes in the original graph to nodes 2575 // in the new graph. 2576 llvm::SmallPtrSet<const ExplodedNode *, 32> RemainingNodes; 2577 2578 for (PathSensitiveBugReport *Report : bugReports) { 2579 const ExplodedNode *NewNode = ForwardMap.lookup(Report->getErrorNode()); 2580 assert(NewNode && 2581 "Failed to construct a trimmed graph that contains this error " 2582 "node!"); 2583 ReportNodes.emplace_back(Report, NewNode); 2584 RemainingNodes.insert(NewNode); 2585 } 2586 2587 assert(!RemainingNodes.empty() && "No error node found in the trimmed graph"); 2588 2589 // Perform a forward BFS to find all the shortest paths. 2590 std::queue<const ExplodedNode *> WS; 2591 2592 assert(TrimmedGraph->num_roots() == 1); 2593 WS.push(*TrimmedGraph->roots_begin()); 2594 unsigned Priority = 0; 2595 2596 while (!WS.empty()) { 2597 const ExplodedNode *Node = WS.front(); 2598 WS.pop(); 2599 2600 PriorityMapTy::iterator PriorityEntry; 2601 bool IsNew; 2602 std::tie(PriorityEntry, IsNew) = PriorityMap.insert({Node, Priority}); 2603 ++Priority; 2604 2605 if (!IsNew) { 2606 assert(PriorityEntry->second <= Priority); 2607 continue; 2608 } 2609 2610 if (RemainingNodes.erase(Node)) 2611 if (RemainingNodes.empty()) 2612 break; 2613 2614 for (const ExplodedNode *Succ : Node->succs()) 2615 WS.push(Succ); 2616 } 2617 2618 // Sort the error paths from longest to shortest. 2619 llvm::sort(ReportNodes, PriorityCompare<true>(PriorityMap)); 2620 } 2621 2622 BugPathInfo *BugPathGetter::getNextBugPath() { 2623 if (ReportNodes.empty()) 2624 return nullptr; 2625 2626 const ExplodedNode *OrigN; 2627 std::tie(CurrentBugPath.Report, OrigN) = ReportNodes.pop_back_val(); 2628 assert(PriorityMap.contains(OrigN) && "error node not accessible from root"); 2629 2630 // Create a new graph with a single path. This is the graph that will be 2631 // returned to the caller. 2632 auto GNew = std::make_unique<ExplodedGraph>(); 2633 2634 // Now walk from the error node up the BFS path, always taking the 2635 // predeccessor with the lowest number. 2636 ExplodedNode *Succ = nullptr; 2637 while (true) { 2638 // Create the equivalent node in the new graph with the same state 2639 // and location. 2640 ExplodedNode *NewN = GNew->createUncachedNode( 2641 OrigN->getLocation(), OrigN->getState(), 2642 OrigN->getID(), OrigN->isSink()); 2643 2644 // Link up the new node with the previous node. 2645 if (Succ) 2646 Succ->addPredecessor(NewN, *GNew); 2647 else 2648 CurrentBugPath.ErrorNode = NewN; 2649 2650 Succ = NewN; 2651 2652 // Are we at the final node? 2653 if (OrigN->pred_empty()) { 2654 GNew->addRoot(NewN); 2655 break; 2656 } 2657 2658 // Find the next predeccessor node. We choose the node that is marked 2659 // with the lowest BFS number. 2660 OrigN = *std::min_element(OrigN->pred_begin(), OrigN->pred_end(), 2661 PriorityCompare<false>(PriorityMap)); 2662 } 2663 2664 CurrentBugPath.BugPath = std::move(GNew); 2665 2666 return &CurrentBugPath; 2667 } 2668 2669 /// CompactMacroExpandedPieces - This function postprocesses a PathDiagnostic 2670 /// object and collapses PathDiagosticPieces that are expanded by macros. 2671 static void CompactMacroExpandedPieces(PathPieces &path, 2672 const SourceManager& SM) { 2673 using MacroStackTy = std::vector< 2674 std::pair<std::shared_ptr<PathDiagnosticMacroPiece>, SourceLocation>>; 2675 2676 using PiecesTy = std::vector<PathDiagnosticPieceRef>; 2677 2678 MacroStackTy MacroStack; 2679 PiecesTy Pieces; 2680 2681 for (PathPieces::const_iterator I = path.begin(), E = path.end(); 2682 I != E; ++I) { 2683 const auto &piece = *I; 2684 2685 // Recursively compact calls. 2686 if (auto *call = dyn_cast<PathDiagnosticCallPiece>(&*piece)) { 2687 CompactMacroExpandedPieces(call->path, SM); 2688 } 2689 2690 // Get the location of the PathDiagnosticPiece. 2691 const FullSourceLoc Loc = piece->getLocation().asLocation(); 2692 2693 // Determine the instantiation location, which is the location we group 2694 // related PathDiagnosticPieces. 2695 SourceLocation InstantiationLoc = Loc.isMacroID() ? 2696 SM.getExpansionLoc(Loc) : 2697 SourceLocation(); 2698 2699 if (Loc.isFileID()) { 2700 MacroStack.clear(); 2701 Pieces.push_back(piece); 2702 continue; 2703 } 2704 2705 assert(Loc.isMacroID()); 2706 2707 // Is the PathDiagnosticPiece within the same macro group? 2708 if (!MacroStack.empty() && InstantiationLoc == MacroStack.back().second) { 2709 MacroStack.back().first->subPieces.push_back(piece); 2710 continue; 2711 } 2712 2713 // We aren't in the same group. Are we descending into a new macro 2714 // or are part of an old one? 2715 std::shared_ptr<PathDiagnosticMacroPiece> MacroGroup; 2716 2717 SourceLocation ParentInstantiationLoc = InstantiationLoc.isMacroID() ? 2718 SM.getExpansionLoc(Loc) : 2719 SourceLocation(); 2720 2721 // Walk the entire macro stack. 2722 while (!MacroStack.empty()) { 2723 if (InstantiationLoc == MacroStack.back().second) { 2724 MacroGroup = MacroStack.back().first; 2725 break; 2726 } 2727 2728 if (ParentInstantiationLoc == MacroStack.back().second) { 2729 MacroGroup = MacroStack.back().first; 2730 break; 2731 } 2732 2733 MacroStack.pop_back(); 2734 } 2735 2736 if (!MacroGroup || ParentInstantiationLoc == MacroStack.back().second) { 2737 // Create a new macro group and add it to the stack. 2738 auto NewGroup = std::make_shared<PathDiagnosticMacroPiece>( 2739 PathDiagnosticLocation::createSingleLocation(piece->getLocation())); 2740 2741 if (MacroGroup) 2742 MacroGroup->subPieces.push_back(NewGroup); 2743 else { 2744 assert(InstantiationLoc.isFileID()); 2745 Pieces.push_back(NewGroup); 2746 } 2747 2748 MacroGroup = NewGroup; 2749 MacroStack.push_back(std::make_pair(MacroGroup, InstantiationLoc)); 2750 } 2751 2752 // Finally, add the PathDiagnosticPiece to the group. 2753 MacroGroup->subPieces.push_back(piece); 2754 } 2755 2756 // Now take the pieces and construct a new PathDiagnostic. 2757 path.clear(); 2758 2759 path.insert(path.end(), Pieces.begin(), Pieces.end()); 2760 } 2761 2762 /// Generate notes from all visitors. 2763 /// Notes associated with @c ErrorNode are generated using 2764 /// @c getEndPath, and the rest are generated with @c VisitNode. 2765 static std::unique_ptr<VisitorsDiagnosticsTy> 2766 generateVisitorsDiagnostics(PathSensitiveBugReport *R, 2767 const ExplodedNode *ErrorNode, 2768 BugReporterContext &BRC) { 2769 std::unique_ptr<VisitorsDiagnosticsTy> Notes = 2770 std::make_unique<VisitorsDiagnosticsTy>(); 2771 PathSensitiveBugReport::VisitorList visitors; 2772 2773 // Run visitors on all nodes starting from the node *before* the last one. 2774 // The last node is reserved for notes generated with @c getEndPath. 2775 const ExplodedNode *NextNode = ErrorNode->getFirstPred(); 2776 while (NextNode) { 2777 2778 // At each iteration, move all visitors from report to visitor list. This is 2779 // important, because the Profile() functions of the visitors make sure that 2780 // a visitor isn't added multiple times for the same node, but it's fine 2781 // to add the a visitor with Profile() for different nodes (e.g. tracking 2782 // a region at different points of the symbolic execution). 2783 for (std::unique_ptr<BugReporterVisitor> &Visitor : R->visitors()) 2784 visitors.push_back(std::move(Visitor)); 2785 2786 R->clearVisitors(); 2787 2788 const ExplodedNode *Pred = NextNode->getFirstPred(); 2789 if (!Pred) { 2790 PathDiagnosticPieceRef LastPiece; 2791 for (auto &V : visitors) { 2792 V->finalizeVisitor(BRC, ErrorNode, *R); 2793 2794 if (auto Piece = V->getEndPath(BRC, ErrorNode, *R)) { 2795 assert(!LastPiece && 2796 "There can only be one final piece in a diagnostic."); 2797 assert(Piece->getKind() == PathDiagnosticPiece::Kind::Event && 2798 "The final piece must contain a message!"); 2799 LastPiece = std::move(Piece); 2800 (*Notes)[ErrorNode].push_back(LastPiece); 2801 } 2802 } 2803 break; 2804 } 2805 2806 for (auto &V : visitors) { 2807 auto P = V->VisitNode(NextNode, BRC, *R); 2808 if (P) 2809 (*Notes)[NextNode].push_back(std::move(P)); 2810 } 2811 2812 if (!R->isValid()) 2813 break; 2814 2815 NextNode = Pred; 2816 } 2817 2818 return Notes; 2819 } 2820 2821 std::optional<PathDiagnosticBuilder> PathDiagnosticBuilder::findValidReport( 2822 ArrayRef<PathSensitiveBugReport *> &bugReports, 2823 PathSensitiveBugReporter &Reporter) { 2824 2825 BugPathGetter BugGraph(&Reporter.getGraph(), bugReports); 2826 2827 while (BugPathInfo *BugPath = BugGraph.getNextBugPath()) { 2828 // Find the BugReport with the original location. 2829 PathSensitiveBugReport *R = BugPath->Report; 2830 assert(R && "No original report found for sliced graph."); 2831 assert(R->isValid() && "Report selected by trimmed graph marked invalid."); 2832 const ExplodedNode *ErrorNode = BugPath->ErrorNode; 2833 2834 // Register refutation visitors first, if they mark the bug invalid no 2835 // further analysis is required 2836 R->addVisitor<LikelyFalsePositiveSuppressionBRVisitor>(); 2837 2838 // Register additional node visitors. 2839 R->addVisitor<NilReceiverBRVisitor>(); 2840 R->addVisitor<ConditionBRVisitor>(); 2841 R->addVisitor<TagVisitor>(); 2842 2843 BugReporterContext BRC(Reporter); 2844 2845 // Run all visitors on a given graph, once. 2846 std::unique_ptr<VisitorsDiagnosticsTy> visitorNotes = 2847 generateVisitorsDiagnostics(R, ErrorNode, BRC); 2848 2849 if (R->isValid()) { 2850 if (Reporter.getAnalyzerOptions().ShouldCrosscheckWithZ3) { 2851 // If crosscheck is enabled, remove all visitors, add the refutation 2852 // visitor and check again 2853 R->clearVisitors(); 2854 R->addVisitor<FalsePositiveRefutationBRVisitor>(); 2855 2856 // We don't overwrite the notes inserted by other visitors because the 2857 // refutation manager does not add any new note to the path 2858 generateVisitorsDiagnostics(R, BugPath->ErrorNode, BRC); 2859 } 2860 2861 // Check if the bug is still valid 2862 if (R->isValid()) 2863 return PathDiagnosticBuilder( 2864 std::move(BRC), std::move(BugPath->BugPath), BugPath->Report, 2865 BugPath->ErrorNode, std::move(visitorNotes)); 2866 } 2867 } 2868 2869 return {}; 2870 } 2871 2872 std::unique_ptr<DiagnosticForConsumerMapTy> 2873 PathSensitiveBugReporter::generatePathDiagnostics( 2874 ArrayRef<PathDiagnosticConsumer *> consumers, 2875 ArrayRef<PathSensitiveBugReport *> &bugReports) { 2876 assert(!bugReports.empty()); 2877 2878 auto Out = std::make_unique<DiagnosticForConsumerMapTy>(); 2879 2880 std::optional<PathDiagnosticBuilder> PDB = 2881 PathDiagnosticBuilder::findValidReport(bugReports, *this); 2882 2883 if (PDB) { 2884 for (PathDiagnosticConsumer *PC : consumers) { 2885 if (std::unique_ptr<PathDiagnostic> PD = PDB->generate(PC)) { 2886 (*Out)[PC] = std::move(PD); 2887 } 2888 } 2889 } 2890 2891 return Out; 2892 } 2893 2894 void BugReporter::emitReport(std::unique_ptr<BugReport> R) { 2895 bool ValidSourceLoc = R->getLocation().isValid(); 2896 assert(ValidSourceLoc); 2897 // If we mess up in a release build, we'd still prefer to just drop the bug 2898 // instead of trying to go on. 2899 if (!ValidSourceLoc) 2900 return; 2901 2902 // Compute the bug report's hash to determine its equivalence class. 2903 llvm::FoldingSetNodeID ID; 2904 R->Profile(ID); 2905 2906 // Lookup the equivance class. If there isn't one, create it. 2907 void *InsertPos; 2908 BugReportEquivClass* EQ = EQClasses.FindNodeOrInsertPos(ID, InsertPos); 2909 2910 if (!EQ) { 2911 EQ = new BugReportEquivClass(std::move(R)); 2912 EQClasses.InsertNode(EQ, InsertPos); 2913 EQClassesVector.push_back(EQ); 2914 } else 2915 EQ->AddReport(std::move(R)); 2916 } 2917 2918 void PathSensitiveBugReporter::emitReport(std::unique_ptr<BugReport> R) { 2919 if (auto PR = dyn_cast<PathSensitiveBugReport>(R.get())) 2920 if (const ExplodedNode *E = PR->getErrorNode()) { 2921 // An error node must either be a sink or have a tag, otherwise 2922 // it could get reclaimed before the path diagnostic is created. 2923 assert((E->isSink() || E->getLocation().getTag()) && 2924 "Error node must either be a sink or have a tag"); 2925 2926 const AnalysisDeclContext *DeclCtx = 2927 E->getLocationContext()->getAnalysisDeclContext(); 2928 // The source of autosynthesized body can be handcrafted AST or a model 2929 // file. The locations from handcrafted ASTs have no valid source 2930 // locations and have to be discarded. Locations from model files should 2931 // be preserved for processing and reporting. 2932 if (DeclCtx->isBodyAutosynthesized() && 2933 !DeclCtx->isBodyAutosynthesizedFromModelFile()) 2934 return; 2935 } 2936 2937 BugReporter::emitReport(std::move(R)); 2938 } 2939 2940 //===----------------------------------------------------------------------===// 2941 // Emitting reports in equivalence classes. 2942 //===----------------------------------------------------------------------===// 2943 2944 namespace { 2945 2946 struct FRIEC_WLItem { 2947 const ExplodedNode *N; 2948 ExplodedNode::const_succ_iterator I, E; 2949 2950 FRIEC_WLItem(const ExplodedNode *n) 2951 : N(n), I(N->succ_begin()), E(N->succ_end()) {} 2952 }; 2953 2954 } // namespace 2955 2956 BugReport *PathSensitiveBugReporter::findReportInEquivalenceClass( 2957 BugReportEquivClass &EQ, SmallVectorImpl<BugReport *> &bugReports) { 2958 // If we don't need to suppress any of the nodes because they are 2959 // post-dominated by a sink, simply add all the nodes in the equivalence class 2960 // to 'Nodes'. Any of the reports will serve as a "representative" report. 2961 assert(EQ.getReports().size() > 0); 2962 const BugType& BT = EQ.getReports()[0]->getBugType(); 2963 if (!BT.isSuppressOnSink()) { 2964 BugReport *R = EQ.getReports()[0].get(); 2965 for (auto &J : EQ.getReports()) { 2966 if (auto *PR = dyn_cast<PathSensitiveBugReport>(J.get())) { 2967 R = PR; 2968 bugReports.push_back(PR); 2969 } 2970 } 2971 return R; 2972 } 2973 2974 // For bug reports that should be suppressed when all paths are post-dominated 2975 // by a sink node, iterate through the reports in the equivalence class 2976 // until we find one that isn't post-dominated (if one exists). We use a 2977 // DFS traversal of the ExplodedGraph to find a non-sink node. We could write 2978 // this as a recursive function, but we don't want to risk blowing out the 2979 // stack for very long paths. 2980 BugReport *exampleReport = nullptr; 2981 2982 for (const auto &I: EQ.getReports()) { 2983 auto *R = dyn_cast<PathSensitiveBugReport>(I.get()); 2984 if (!R) 2985 continue; 2986 2987 const ExplodedNode *errorNode = R->getErrorNode(); 2988 if (errorNode->isSink()) { 2989 llvm_unreachable( 2990 "BugType::isSuppressSink() should not be 'true' for sink end nodes"); 2991 } 2992 // No successors? By definition this nodes isn't post-dominated by a sink. 2993 if (errorNode->succ_empty()) { 2994 bugReports.push_back(R); 2995 if (!exampleReport) 2996 exampleReport = R; 2997 continue; 2998 } 2999 3000 // See if we are in a no-return CFG block. If so, treat this similarly 3001 // to being post-dominated by a sink. This works better when the analysis 3002 // is incomplete and we have never reached the no-return function call(s) 3003 // that we'd inevitably bump into on this path. 3004 if (const CFGBlock *ErrorB = errorNode->getCFGBlock()) 3005 if (ErrorB->isInevitablySinking()) 3006 continue; 3007 3008 // At this point we know that 'N' is not a sink and it has at least one 3009 // successor. Use a DFS worklist to find a non-sink end-of-path node. 3010 using WLItem = FRIEC_WLItem; 3011 using DFSWorkList = SmallVector<WLItem, 10>; 3012 3013 llvm::DenseMap<const ExplodedNode *, unsigned> Visited; 3014 3015 DFSWorkList WL; 3016 WL.push_back(errorNode); 3017 Visited[errorNode] = 1; 3018 3019 while (!WL.empty()) { 3020 WLItem &WI = WL.back(); 3021 assert(!WI.N->succ_empty()); 3022 3023 for (; WI.I != WI.E; ++WI.I) { 3024 const ExplodedNode *Succ = *WI.I; 3025 // End-of-path node? 3026 if (Succ->succ_empty()) { 3027 // If we found an end-of-path node that is not a sink. 3028 if (!Succ->isSink()) { 3029 bugReports.push_back(R); 3030 if (!exampleReport) 3031 exampleReport = R; 3032 WL.clear(); 3033 break; 3034 } 3035 // Found a sink? Continue on to the next successor. 3036 continue; 3037 } 3038 // Mark the successor as visited. If it hasn't been explored, 3039 // enqueue it to the DFS worklist. 3040 unsigned &mark = Visited[Succ]; 3041 if (!mark) { 3042 mark = 1; 3043 WL.push_back(Succ); 3044 break; 3045 } 3046 } 3047 3048 // The worklist may have been cleared at this point. First 3049 // check if it is empty before checking the last item. 3050 if (!WL.empty() && &WL.back() == &WI) 3051 WL.pop_back(); 3052 } 3053 } 3054 3055 // ExampleReport will be NULL if all the nodes in the equivalence class 3056 // were post-dominated by sinks. 3057 return exampleReport; 3058 } 3059 3060 void BugReporter::FlushReport(BugReportEquivClass& EQ) { 3061 SmallVector<BugReport*, 10> bugReports; 3062 BugReport *report = findReportInEquivalenceClass(EQ, bugReports); 3063 if (!report) 3064 return; 3065 3066 // See whether we need to silence the checker/package. 3067 for (const std::string &CheckerOrPackage : 3068 getAnalyzerOptions().SilencedCheckersAndPackages) { 3069 if (report->getBugType().getCheckerName().startswith( 3070 CheckerOrPackage)) 3071 return; 3072 } 3073 3074 ArrayRef<PathDiagnosticConsumer*> Consumers = getPathDiagnosticConsumers(); 3075 std::unique_ptr<DiagnosticForConsumerMapTy> Diagnostics = 3076 generateDiagnosticForConsumerMap(report, Consumers, bugReports); 3077 3078 for (auto &P : *Diagnostics) { 3079 PathDiagnosticConsumer *Consumer = P.first; 3080 std::unique_ptr<PathDiagnostic> &PD = P.second; 3081 3082 // If the path is empty, generate a single step path with the location 3083 // of the issue. 3084 if (PD->path.empty()) { 3085 PathDiagnosticLocation L = report->getLocation(); 3086 auto piece = std::make_unique<PathDiagnosticEventPiece>( 3087 L, report->getDescription()); 3088 for (SourceRange Range : report->getRanges()) 3089 piece->addRange(Range); 3090 PD->setEndOfPath(std::move(piece)); 3091 } 3092 3093 PathPieces &Pieces = PD->getMutablePieces(); 3094 if (getAnalyzerOptions().ShouldDisplayNotesAsEvents) { 3095 // For path diagnostic consumers that don't support extra notes, 3096 // we may optionally convert those to path notes. 3097 for (const auto &I : llvm::reverse(report->getNotes())) { 3098 PathDiagnosticNotePiece *Piece = I.get(); 3099 auto ConvertedPiece = std::make_shared<PathDiagnosticEventPiece>( 3100 Piece->getLocation(), Piece->getString()); 3101 for (const auto &R: Piece->getRanges()) 3102 ConvertedPiece->addRange(R); 3103 3104 Pieces.push_front(std::move(ConvertedPiece)); 3105 } 3106 } else { 3107 for (const auto &I : llvm::reverse(report->getNotes())) 3108 Pieces.push_front(I); 3109 } 3110 3111 for (const auto &I : report->getFixits()) 3112 Pieces.back()->addFixit(I); 3113 3114 updateExecutedLinesWithDiagnosticPieces(*PD); 3115 Consumer->HandlePathDiagnostic(std::move(PD)); 3116 } 3117 } 3118 3119 /// Insert all lines participating in the function signature \p Signature 3120 /// into \p ExecutedLines. 3121 static void populateExecutedLinesWithFunctionSignature( 3122 const Decl *Signature, const SourceManager &SM, 3123 FilesToLineNumsMap &ExecutedLines) { 3124 SourceRange SignatureSourceRange; 3125 const Stmt* Body = Signature->getBody(); 3126 if (const auto FD = dyn_cast<FunctionDecl>(Signature)) { 3127 SignatureSourceRange = FD->getSourceRange(); 3128 } else if (const auto OD = dyn_cast<ObjCMethodDecl>(Signature)) { 3129 SignatureSourceRange = OD->getSourceRange(); 3130 } else { 3131 return; 3132 } 3133 SourceLocation Start = SignatureSourceRange.getBegin(); 3134 SourceLocation End = Body ? Body->getSourceRange().getBegin() 3135 : SignatureSourceRange.getEnd(); 3136 if (!Start.isValid() || !End.isValid()) 3137 return; 3138 unsigned StartLine = SM.getExpansionLineNumber(Start); 3139 unsigned EndLine = SM.getExpansionLineNumber(End); 3140 3141 FileID FID = SM.getFileID(SM.getExpansionLoc(Start)); 3142 for (unsigned Line = StartLine; Line <= EndLine; Line++) 3143 ExecutedLines[FID].insert(Line); 3144 } 3145 3146 static void populateExecutedLinesWithStmt( 3147 const Stmt *S, const SourceManager &SM, 3148 FilesToLineNumsMap &ExecutedLines) { 3149 SourceLocation Loc = S->getSourceRange().getBegin(); 3150 if (!Loc.isValid()) 3151 return; 3152 SourceLocation ExpansionLoc = SM.getExpansionLoc(Loc); 3153 FileID FID = SM.getFileID(ExpansionLoc); 3154 unsigned LineNo = SM.getExpansionLineNumber(ExpansionLoc); 3155 ExecutedLines[FID].insert(LineNo); 3156 } 3157 3158 /// \return all executed lines including function signatures on the path 3159 /// starting from \p N. 3160 static std::unique_ptr<FilesToLineNumsMap> 3161 findExecutedLines(const SourceManager &SM, const ExplodedNode *N) { 3162 auto ExecutedLines = std::make_unique<FilesToLineNumsMap>(); 3163 3164 while (N) { 3165 if (N->getFirstPred() == nullptr) { 3166 // First node: show signature of the entrance point. 3167 const Decl *D = N->getLocationContext()->getDecl(); 3168 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines); 3169 } else if (auto CE = N->getLocationAs<CallEnter>()) { 3170 // Inlined function: show signature. 3171 const Decl* D = CE->getCalleeContext()->getDecl(); 3172 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines); 3173 } else if (const Stmt *S = N->getStmtForDiagnostics()) { 3174 populateExecutedLinesWithStmt(S, SM, *ExecutedLines); 3175 3176 // Show extra context for some parent kinds. 3177 const Stmt *P = N->getParentMap().getParent(S); 3178 3179 // The path exploration can die before the node with the associated 3180 // return statement is generated, but we do want to show the whole 3181 // return. 3182 if (const auto *RS = dyn_cast_or_null<ReturnStmt>(P)) { 3183 populateExecutedLinesWithStmt(RS, SM, *ExecutedLines); 3184 P = N->getParentMap().getParent(RS); 3185 } 3186 3187 if (isa_and_nonnull<SwitchCase, LabelStmt>(P)) 3188 populateExecutedLinesWithStmt(P, SM, *ExecutedLines); 3189 } 3190 3191 N = N->getFirstPred(); 3192 } 3193 return ExecutedLines; 3194 } 3195 3196 std::unique_ptr<DiagnosticForConsumerMapTy> 3197 BugReporter::generateDiagnosticForConsumerMap( 3198 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers, 3199 ArrayRef<BugReport *> bugReports) { 3200 auto *basicReport = cast<BasicBugReport>(exampleReport); 3201 auto Out = std::make_unique<DiagnosticForConsumerMapTy>(); 3202 for (auto *Consumer : consumers) 3203 (*Out)[Consumer] = generateDiagnosticForBasicReport(basicReport); 3204 return Out; 3205 } 3206 3207 static PathDiagnosticCallPiece * 3208 getFirstStackedCallToHeaderFile(PathDiagnosticCallPiece *CP, 3209 const SourceManager &SMgr) { 3210 SourceLocation CallLoc = CP->callEnter.asLocation(); 3211 3212 // If the call is within a macro, don't do anything (for now). 3213 if (CallLoc.isMacroID()) 3214 return nullptr; 3215 3216 assert(AnalysisManager::isInCodeFile(CallLoc, SMgr) && 3217 "The call piece should not be in a header file."); 3218 3219 // Check if CP represents a path through a function outside of the main file. 3220 if (!AnalysisManager::isInCodeFile(CP->callEnterWithin.asLocation(), SMgr)) 3221 return CP; 3222 3223 const PathPieces &Path = CP->path; 3224 if (Path.empty()) 3225 return nullptr; 3226 3227 // Check if the last piece in the callee path is a call to a function outside 3228 // of the main file. 3229 if (auto *CPInner = dyn_cast<PathDiagnosticCallPiece>(Path.back().get())) 3230 return getFirstStackedCallToHeaderFile(CPInner, SMgr); 3231 3232 // Otherwise, the last piece is in the main file. 3233 return nullptr; 3234 } 3235 3236 static void resetDiagnosticLocationToMainFile(PathDiagnostic &PD) { 3237 if (PD.path.empty()) 3238 return; 3239 3240 PathDiagnosticPiece *LastP = PD.path.back().get(); 3241 assert(LastP); 3242 const SourceManager &SMgr = LastP->getLocation().getManager(); 3243 3244 // We only need to check if the report ends inside headers, if the last piece 3245 // is a call piece. 3246 if (auto *CP = dyn_cast<PathDiagnosticCallPiece>(LastP)) { 3247 CP = getFirstStackedCallToHeaderFile(CP, SMgr); 3248 if (CP) { 3249 // Mark the piece. 3250 CP->setAsLastInMainSourceFile(); 3251 3252 // Update the path diagnostic message. 3253 const auto *ND = dyn_cast<NamedDecl>(CP->getCallee()); 3254 if (ND) { 3255 SmallString<200> buf; 3256 llvm::raw_svector_ostream os(buf); 3257 os << " (within a call to '" << ND->getDeclName() << "')"; 3258 PD.appendToDesc(os.str()); 3259 } 3260 3261 // Reset the report containing declaration and location. 3262 PD.setDeclWithIssue(CP->getCaller()); 3263 PD.setLocation(CP->getLocation()); 3264 3265 return; 3266 } 3267 } 3268 } 3269 3270 3271 3272 std::unique_ptr<DiagnosticForConsumerMapTy> 3273 PathSensitiveBugReporter::generateDiagnosticForConsumerMap( 3274 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers, 3275 ArrayRef<BugReport *> bugReports) { 3276 std::vector<BasicBugReport *> BasicBugReports; 3277 std::vector<PathSensitiveBugReport *> PathSensitiveBugReports; 3278 if (isa<BasicBugReport>(exampleReport)) 3279 return BugReporter::generateDiagnosticForConsumerMap(exampleReport, 3280 consumers, bugReports); 3281 3282 // Generate the full path sensitive diagnostic, using the generation scheme 3283 // specified by the PathDiagnosticConsumer. Note that we have to generate 3284 // path diagnostics even for consumers which do not support paths, because 3285 // the BugReporterVisitors may mark this bug as a false positive. 3286 assert(!bugReports.empty()); 3287 MaxBugClassSize.updateMax(bugReports.size()); 3288 3289 // Avoid copying the whole array because there may be a lot of reports. 3290 ArrayRef<PathSensitiveBugReport *> convertedArrayOfReports( 3291 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.begin()), 3292 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.end())); 3293 std::unique_ptr<DiagnosticForConsumerMapTy> Out = generatePathDiagnostics( 3294 consumers, convertedArrayOfReports); 3295 3296 if (Out->empty()) 3297 return Out; 3298 3299 MaxValidBugClassSize.updateMax(bugReports.size()); 3300 3301 // Examine the report and see if the last piece is in a header. Reset the 3302 // report location to the last piece in the main source file. 3303 const AnalyzerOptions &Opts = getAnalyzerOptions(); 3304 for (auto const &P : *Out) 3305 if (Opts.ShouldReportIssuesInMainSourceFile && !Opts.AnalyzeAll) 3306 resetDiagnosticLocationToMainFile(*P.second); 3307 3308 return Out; 3309 } 3310 3311 void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, 3312 const CheckerBase *Checker, StringRef Name, 3313 StringRef Category, StringRef Str, 3314 PathDiagnosticLocation Loc, 3315 ArrayRef<SourceRange> Ranges, 3316 ArrayRef<FixItHint> Fixits) { 3317 EmitBasicReport(DeclWithIssue, Checker->getCheckerName(), Name, Category, Str, 3318 Loc, Ranges, Fixits); 3319 } 3320 3321 void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, 3322 CheckerNameRef CheckName, 3323 StringRef name, StringRef category, 3324 StringRef str, PathDiagnosticLocation Loc, 3325 ArrayRef<SourceRange> Ranges, 3326 ArrayRef<FixItHint> Fixits) { 3327 // 'BT' is owned by BugReporter. 3328 BugType *BT = getBugTypeForName(CheckName, name, category); 3329 auto R = std::make_unique<BasicBugReport>(*BT, str, Loc); 3330 R->setDeclWithIssue(DeclWithIssue); 3331 for (const auto &SR : Ranges) 3332 R->addRange(SR); 3333 for (const auto &FH : Fixits) 3334 R->addFixItHint(FH); 3335 emitReport(std::move(R)); 3336 } 3337 3338 BugType *BugReporter::getBugTypeForName(CheckerNameRef CheckName, 3339 StringRef name, StringRef category) { 3340 SmallString<136> fullDesc; 3341 llvm::raw_svector_ostream(fullDesc) << CheckName.getName() << ":" << name 3342 << ":" << category; 3343 std::unique_ptr<BugType> &BT = StrBugTypes[fullDesc]; 3344 if (!BT) 3345 BT = std::make_unique<BugType>(CheckName, name, category); 3346 return BT.get(); 3347 } 3348