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/None.h"
50#include "llvm/ADT/Optional.h"
51#include "llvm/ADT/STLExtras.h"
52#include "llvm/ADT/SmallPtrSet.h"
53#include "llvm/ADT/SmallString.h"
54#include "llvm/ADT/SmallVector.h"
55#include "llvm/ADT/Statistic.h"
56#include "llvm/ADT/StringExtras.h"
57#include "llvm/ADT/StringRef.h"
58#include "llvm/ADT/iterator_range.h"
59#include "llvm/Support/Casting.h"
60#include "llvm/Support/Compiler.h"
61#include "llvm/Support/ErrorHandling.h"
62#include "llvm/Support/MemoryBuffer.h"
63#include "llvm/Support/raw_ostream.h"
64#include <algorithm>
65#include <cassert>
66#include <cstddef>
67#include <iterator>
68#include <memory>
69#include <queue>
70#include <string>
71#include <tuple>
72#include <utility>
73#include <vector>
74
75using namespace clang;
76using namespace ento;
77using namespace llvm;
78
79#define DEBUG_TYPE "BugReporter"
80
81STATISTIC(MaxBugClassSize,
82          "The maximum number of bug reports in the same equivalence class");
83STATISTIC(MaxValidBugClassSize,
84          "The maximum number of bug reports in the same equivalence class "
85          "where at least one report is valid (not suppressed)");
86
87BugReporterVisitor::~BugReporterVisitor() = default;
88
89void BugReporterContext::anchor() {}
90
91//===----------------------------------------------------------------------===//
92// PathDiagnosticBuilder and its associated routines and helper objects.
93//===----------------------------------------------------------------------===//
94
95namespace {
96
97/// A (CallPiece, node assiciated with its CallEnter) pair.
98using CallWithEntry =
99    std::pair<PathDiagnosticCallPiece *, const ExplodedNode *>;
100using CallWithEntryStack = SmallVector<CallWithEntry, 6>;
101
102/// Map from each node to the diagnostic pieces visitors emit for them.
103using VisitorsDiagnosticsTy =
104    llvm::DenseMap<const ExplodedNode *, std::vector<PathDiagnosticPieceRef>>;
105
106/// A map from PathDiagnosticPiece to the LocationContext of the inlined
107/// function call it represents.
108using LocationContextMap =
109    llvm::DenseMap<const PathPieces *, const LocationContext *>;
110
111/// A helper class that contains everything needed to construct a
112/// PathDiagnostic object. It does no much more then providing convenient
113/// getters and some well placed asserts for extra security.
114class PathDiagnosticConstruct {
115  /// The consumer we're constructing the bug report for.
116  const PathDiagnosticConsumer *Consumer;
117  /// Our current position in the bug path, which is owned by
118  /// PathDiagnosticBuilder.
119  const ExplodedNode *CurrentNode;
120  /// A mapping from parts of the bug path (for example, a function call, which
121  /// would span backwards from a CallExit to a CallEnter with the nodes in
122  /// between them) with the location contexts it is associated with.
123  LocationContextMap LCM;
124  const SourceManager &SM;
125
126public:
127  /// We keep stack of calls to functions as we're ascending the bug path.
128  /// TODO: PathDiagnostic has a stack doing the same thing, shouldn't we use
129  /// that instead?
130  CallWithEntryStack CallStack;
131  /// The bug report we're constructing. For ease of use, this field is kept
132  /// public, though some "shortcut" getters are provided for commonly used
133  /// methods of PathDiagnostic.
134  std::unique_ptr<PathDiagnostic> PD;
135
136public:
137  PathDiagnosticConstruct(const PathDiagnosticConsumer *PDC,
138                          const ExplodedNode *ErrorNode,
139                          const PathSensitiveBugReport *R);
140
141  /// \returns the location context associated with the current position in the
142  /// bug path.
143  const LocationContext *getCurrLocationContext() const {
144    assert(CurrentNode && "Already reached the root!");
145    return CurrentNode->getLocationContext();
146  }
147
148  /// Same as getCurrLocationContext (they should always return the same
149  /// location context), but works after reaching the root of the bug path as
150  /// well.
151  const LocationContext *getLocationContextForActivePath() const {
152    return LCM.find(&PD->getActivePath())->getSecond();
153  }
154
155  const ExplodedNode *getCurrentNode() const { return CurrentNode; }
156
157  /// Steps the current node to its predecessor.
158  /// \returns whether we reached the root of the bug path.
159  bool ascendToPrevNode() {
160    CurrentNode = CurrentNode->getFirstPred();
161    return static_cast<bool>(CurrentNode);
162  }
163
164  const ParentMap &getParentMap() const {
165    return getCurrLocationContext()->getParentMap();
166  }
167
168  const SourceManager &getSourceManager() const { return SM; }
169
170  const Stmt *getParent(const Stmt *S) const {
171    return getParentMap().getParent(S);
172  }
173
174  void updateLocCtxMap(const PathPieces *Path, const LocationContext *LC) {
175    assert(Path && LC);
176    LCM[Path] = LC;
177  }
178
179  const LocationContext *getLocationContextFor(const PathPieces *Path) const {
180    assert(LCM.count(Path) &&
181           "Failed to find the context associated with these pieces!");
182    return LCM.find(Path)->getSecond();
183  }
184
185  bool isInLocCtxMap(const PathPieces *Path) const { return LCM.count(Path); }
186
187  PathPieces &getActivePath() { return PD->getActivePath(); }
188  PathPieces &getMutablePieces() { return PD->getMutablePieces(); }
189
190  bool shouldAddPathEdges() const { return Consumer->shouldAddPathEdges(); }
191  bool shouldGenerateDiagnostics() const {
192    return Consumer->shouldGenerateDiagnostics();
193  }
194  bool supportsLogicalOpControlFlow() const {
195    return Consumer->supportsLogicalOpControlFlow();
196  }
197};
198
199/// Contains every contextual information needed for constructing a
200/// PathDiagnostic object for a given bug report. This class and its fields are
201/// immutable, and passes a BugReportConstruct object around during the
202/// construction.
203class PathDiagnosticBuilder : public BugReporterContext {
204  /// A linear path from the error node to the root.
205  std::unique_ptr<const ExplodedGraph> BugPath;
206  /// The bug report we're describing. Visitors create their diagnostics with
207  /// them being the last entities being able to modify it (for example,
208  /// changing interestingness here would cause inconsistencies as to how this
209  /// file and visitors construct diagnostics), hence its const.
210  const PathSensitiveBugReport *R;
211  /// The leaf of the bug path. This isn't the same as the bug reports error
212  /// node, which refers to the *original* graph, not the bug path.
213  const ExplodedNode *const ErrorNode;
214  /// The diagnostic pieces visitors emitted, which is expected to be collected
215  /// by the time this builder is constructed.
216  std::unique_ptr<const VisitorsDiagnosticsTy> VisitorsDiagnostics;
217
218public:
219  /// Find a non-invalidated report for a given equivalence class,  and returns
220  /// a PathDiagnosticBuilder able to construct bug reports for different
221  /// consumers. Returns None if no valid report is found.
222  static Optional<PathDiagnosticBuilder>
223  findValidReport(ArrayRef<PathSensitiveBugReport *> &bugReports,
224                  PathSensitiveBugReporter &Reporter);
225
226  PathDiagnosticBuilder(
227      BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath,
228      PathSensitiveBugReport *r, const ExplodedNode *ErrorNode,
229      std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics);
230
231  /// This function is responsible for generating diagnostic pieces that are
232  /// *not* provided by bug report visitors.
233  /// These diagnostics may differ depending on the consumer's settings,
234  /// and are therefore constructed separately for each consumer.
235  ///
236  /// There are two path diagnostics generation modes: with adding edges (used
237  /// for plists) and without  (used for HTML and text). When edges are added,
238  /// the path is modified to insert artificially generated edges.
239  /// Otherwise, more detailed diagnostics is emitted for block edges,
240  /// explaining the transitions in words.
241  std::unique_ptr<PathDiagnostic>
242  generate(const PathDiagnosticConsumer *PDC) const;
243
244private:
245  void updateStackPiecesWithMessage(PathDiagnosticPieceRef P,
246                                    const CallWithEntryStack &CallStack) const;
247  void generatePathDiagnosticsForNode(PathDiagnosticConstruct &C,
248                                      PathDiagnosticLocation &PrevLoc) const;
249
250  void generateMinimalDiagForBlockEdge(PathDiagnosticConstruct &C,
251                                       BlockEdge BE) const;
252
253  PathDiagnosticPieceRef
254  generateDiagForGotoOP(const PathDiagnosticConstruct &C, const Stmt *S,
255                        PathDiagnosticLocation &Start) const;
256
257  PathDiagnosticPieceRef
258  generateDiagForSwitchOP(const PathDiagnosticConstruct &C, const CFGBlock *Dst,
259                          PathDiagnosticLocation &Start) const;
260
261  PathDiagnosticPieceRef
262  generateDiagForBinaryOP(const PathDiagnosticConstruct &C, const Stmt *T,
263                          const CFGBlock *Src, const CFGBlock *DstC) const;
264
265  PathDiagnosticLocation
266  ExecutionContinues(const PathDiagnosticConstruct &C) const;
267
268  PathDiagnosticLocation
269  ExecutionContinues(llvm::raw_string_ostream &os,
270                     const PathDiagnosticConstruct &C) const;
271
272  const PathSensitiveBugReport *getBugReport() const { return R; }
273};
274
275} // namespace
276
277//===----------------------------------------------------------------------===//
278// Base implementation of stack hint generators.
279//===----------------------------------------------------------------------===//
280
281StackHintGenerator::~StackHintGenerator() = default;
282
283std::string StackHintGeneratorForSymbol::getMessage(const ExplodedNode *N){
284  if (!N)
285    return getMessageForSymbolNotFound();
286
287  ProgramPoint P = N->getLocation();
288  CallExitEnd CExit = P.castAs<CallExitEnd>();
289
290  // FIXME: Use CallEvent to abstract this over all calls.
291  const Stmt *CallSite = CExit.getCalleeContext()->getCallSite();
292  const auto *CE = dyn_cast_or_null<CallExpr>(CallSite);
293  if (!CE)
294    return {};
295
296  // Check if one of the parameters are set to the interesting symbol.
297  unsigned ArgIndex = 0;
298  for (CallExpr::const_arg_iterator I = CE->arg_begin(),
299                                    E = CE->arg_end(); I != E; ++I, ++ArgIndex){
300    SVal SV = N->getSVal(*I);
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(*I, ArgIndex);
306    }
307
308    // Check if the parameter is a pointer to the symbol.
309    if (Optional<loc::MemRegionVal> Reg = SV.getAs<loc::MemRegionVal>()) {
310      // Do not attempt to dereference void*.
311      if ((*I)->getType()->isVoidPointerType())
312        continue;
313      SVal PSV = N->getState()->getSVal(Reg->getRegion());
314      SymbolRef AS = PSV.getAsLocSymbol();
315      if (AS == Sym) {
316        return getMessageForArg(*I, ArgIndex);
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
331std::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
344static PathDiagnosticEventPiece *
345eventsDescribeSameCondition(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.
371static 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.
421static 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 &macro = 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.
474static 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.
485static 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.
492static void
493adjustCallLocations(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.
526static 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 (Start && isa<CXXDefaultInitExpr>(Start)) {
538        I = Pieces.erase(I);
539        continue;
540      } else if (End && isa<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.
560static 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
577PathDiagnosticLocation 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
587PathDiagnosticLocation 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
616static 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
638static PathDiagnosticLocation
639getEnclosingStmtLocation(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
721void 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
737static void CompactMacroExpandedPieces(PathPieces &path,
738                                       const SourceManager& SM);
739
740PathDiagnosticPieceRef 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()->IgnoreParenCasts();
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
798PathDiagnosticPieceRef 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
809PathDiagnosticPieceRef 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
857void 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
1001static 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
1014static 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
1020static 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
1030static const Stmt *getStmtBeforeCond(const ParentMap &PM, const Stmt *Term,
1031                                     const ExplodedNode *N) {
1032  while (N) {
1033    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
1044static 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.
1078static 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.
1105static 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
1112constexpr llvm::StringLiteral StrEnteringLoop = "Entering loop body";
1113constexpr llvm::StringLiteral StrLoopBodyZero = "Loop body executed 0 times";
1114constexpr llvm::StringLiteral StrLoopRangeEmpty =
1115    "Loop body skipped when range is empty";
1116constexpr llvm::StringLiteral StrLoopCollectionEmpty =
1117    "Loop body skipped when collection is empty";
1118
1119static std::unique_ptr<FilesToLineNumsMap>
1120findExecutedLines(const SourceManager &SM, const ExplodedNode *N);
1121
1122void 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 (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.shouldAddPathEdges()) {
1236      generateMinimalDiagForBlockEdge(C, *BE);
1237      return;
1238    }
1239
1240    // Are we jumping to the head of a loop?  Add a special diagnostic.
1241    if (const Stmt *Loop = BE->getSrc()->getLoopTarget()) {
1242      PathDiagnosticLocation L(Loop, SM, C.getCurrLocationContext());
1243      const Stmt *Body = nullptr;
1244
1245      if (const auto *FS = dyn_cast<ForStmt>(Loop))
1246        Body = FS->getBody();
1247      else if (const auto *WS = dyn_cast<WhileStmt>(Loop))
1248        Body = WS->getBody();
1249      else if (const auto *OFS = dyn_cast<ObjCForCollectionStmt>(Loop)) {
1250        Body = OFS->getBody();
1251      } else if (const auto *FRS = dyn_cast<CXXForRangeStmt>(Loop)) {
1252        Body = FRS->getBody();
1253      }
1254      // do-while statements are explicitly excluded here
1255
1256      auto p = std::make_shared<PathDiagnosticEventPiece>(
1257          L, "Looping back to the head "
1258          "of the loop");
1259      p->setPrunable(true);
1260
1261      addEdgeToPath(C.getActivePath(), PrevLoc, p->getLocation());
1262      C.getActivePath().push_front(std::move(p));
1263
1264      if (const auto *CS = dyn_cast_or_null<CompoundStmt>(Body)) {
1265        addEdgeToPath(C.getActivePath(), PrevLoc,
1266                      PathDiagnosticLocation::createEndBrace(CS, SM));
1267      }
1268    }
1269
1270    const CFGBlock *BSrc = BE->getSrc();
1271    const ParentMap &PM = C.getParentMap();
1272
1273    if (const Stmt *Term = BSrc->getTerminatorStmt()) {
1274      // Are we jumping past the loop body without ever executing the
1275      // loop (because the condition was false)?
1276      if (isLoop(Term)) {
1277        const Stmt *TermCond = getTerminatorCondition(BSrc);
1278        bool IsInLoopBody = isInLoopBody(
1279            PM, getStmtBeforeCond(PM, TermCond, C.getCurrentNode()), Term);
1280
1281        StringRef str;
1282
1283        if (isJumpToFalseBranch(&*BE)) {
1284          if (!IsInLoopBody) {
1285            if (isa<ObjCForCollectionStmt>(Term)) {
1286              str = StrLoopCollectionEmpty;
1287            } else if (isa<CXXForRangeStmt>(Term)) {
1288              str = StrLoopRangeEmpty;
1289            } else {
1290              str = StrLoopBodyZero;
1291            }
1292          }
1293        } else {
1294          str = StrEnteringLoop;
1295        }
1296
1297        if (!str.empty()) {
1298          PathDiagnosticLocation L(TermCond ? TermCond : Term, SM,
1299                                   C.getCurrLocationContext());
1300          auto PE = std::make_shared<PathDiagnosticEventPiece>(L, str);
1301          PE->setPrunable(true);
1302          addEdgeToPath(C.getActivePath(), PrevLoc, PE->getLocation());
1303          C.getActivePath().push_front(std::move(PE));
1304        }
1305      } else if (isa<BreakStmt>(Term) || isa<ContinueStmt>(Term) ||
1306          isa<GotoStmt>(Term)) {
1307        PathDiagnosticLocation L(Term, SM, C.getCurrLocationContext());
1308        addEdgeToPath(C.getActivePath(), PrevLoc, L);
1309      }
1310    }
1311  }
1312}
1313
1314static std::unique_ptr<PathDiagnostic>
1315generateDiagnosticForBasicReport(const BasicBugReport *R) {
1316  const BugType &BT = R->getBugType();
1317  return std::make_unique<PathDiagnostic>(
1318      BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(),
1319      R->getDescription(), R->getShortDescription(/*UseFallback=*/false),
1320      BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(),
1321      std::make_unique<FilesToLineNumsMap>());
1322}
1323
1324static std::unique_ptr<PathDiagnostic>
1325generateEmptyDiagnosticForReport(const PathSensitiveBugReport *R,
1326                                 const SourceManager &SM) {
1327  const BugType &BT = R->getBugType();
1328  return std::make_unique<PathDiagnostic>(
1329      BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(),
1330      R->getDescription(), R->getShortDescription(/*UseFallback=*/false),
1331      BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(),
1332      findExecutedLines(SM, R->getErrorNode()));
1333}
1334
1335static const Stmt *getStmtParent(const Stmt *S, const ParentMap &PM) {
1336  if (!S)
1337    return nullptr;
1338
1339  while (true) {
1340    S = PM.getParentIgnoreParens(S);
1341
1342    if (!S)
1343      break;
1344
1345    if (isa<FullExpr>(S) ||
1346        isa<CXXBindTemporaryExpr>(S) ||
1347        isa<SubstNonTypeTemplateParmExpr>(S))
1348      continue;
1349
1350    break;
1351  }
1352
1353  return S;
1354}
1355
1356static bool isConditionForTerminator(const Stmt *S, const Stmt *Cond) {
1357  switch (S->getStmtClass()) {
1358    case Stmt::BinaryOperatorClass: {
1359      const auto *BO = cast<BinaryOperator>(S);
1360      if (!BO->isLogicalOp())
1361        return false;
1362      return BO->getLHS() == Cond || BO->getRHS() == Cond;
1363    }
1364    case Stmt::IfStmtClass:
1365      return cast<IfStmt>(S)->getCond() == Cond;
1366    case Stmt::ForStmtClass:
1367      return cast<ForStmt>(S)->getCond() == Cond;
1368    case Stmt::WhileStmtClass:
1369      return cast<WhileStmt>(S)->getCond() == Cond;
1370    case Stmt::DoStmtClass:
1371      return cast<DoStmt>(S)->getCond() == Cond;
1372    case Stmt::ChooseExprClass:
1373      return cast<ChooseExpr>(S)->getCond() == Cond;
1374    case Stmt::IndirectGotoStmtClass:
1375      return cast<IndirectGotoStmt>(S)->getTarget() == Cond;
1376    case Stmt::SwitchStmtClass:
1377      return cast<SwitchStmt>(S)->getCond() == Cond;
1378    case Stmt::BinaryConditionalOperatorClass:
1379      return cast<BinaryConditionalOperator>(S)->getCond() == Cond;
1380    case Stmt::ConditionalOperatorClass: {
1381      const auto *CO = cast<ConditionalOperator>(S);
1382      return CO->getCond() == Cond ||
1383             CO->getLHS() == Cond ||
1384             CO->getRHS() == Cond;
1385    }
1386    case Stmt::ObjCForCollectionStmtClass:
1387      return cast<ObjCForCollectionStmt>(S)->getElement() == Cond;
1388    case Stmt::CXXForRangeStmtClass: {
1389      const auto *FRS = cast<CXXForRangeStmt>(S);
1390      return FRS->getCond() == Cond || FRS->getRangeInit() == Cond;
1391    }
1392    default:
1393      return false;
1394  }
1395}
1396
1397static bool isIncrementOrInitInForLoop(const Stmt *S, const Stmt *FL) {
1398  if (const auto *FS = dyn_cast<ForStmt>(FL))
1399    return FS->getInc() == S || FS->getInit() == S;
1400  if (const auto *FRS = dyn_cast<CXXForRangeStmt>(FL))
1401    return FRS->getInc() == S || FRS->getRangeStmt() == S ||
1402           FRS->getLoopVarStmt() || FRS->getRangeInit() == S;
1403  return false;
1404}
1405
1406using OptimizedCallsSet = llvm::DenseSet<const PathDiagnosticCallPiece *>;
1407
1408/// Adds synthetic edges from top-level statements to their subexpressions.
1409///
1410/// This avoids a "swoosh" effect, where an edge from a top-level statement A
1411/// points to a sub-expression B.1 that's not at the start of B. In these cases,
1412/// we'd like to see an edge from A to B, then another one from B to B.1.
1413static void addContextEdges(PathPieces &pieces, const LocationContext *LC) {
1414  const ParentMap &PM = LC->getParentMap();
1415  PathPieces::iterator Prev = pieces.end();
1416  for (PathPieces::iterator I = pieces.begin(), E = Prev; I != E;
1417       Prev = I, ++I) {
1418    auto *Piece = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1419
1420    if (!Piece)
1421      continue;
1422
1423    PathDiagnosticLocation SrcLoc = Piece->getStartLocation();
1424    SmallVector<PathDiagnosticLocation, 4> SrcContexts;
1425
1426    PathDiagnosticLocation NextSrcContext = SrcLoc;
1427    const Stmt *InnerStmt = nullptr;
1428    while (NextSrcContext.isValid() && NextSrcContext.asStmt() != InnerStmt) {
1429      SrcContexts.push_back(NextSrcContext);
1430      InnerStmt = NextSrcContext.asStmt();
1431      NextSrcContext = getEnclosingStmtLocation(InnerStmt, LC,
1432                                                /*allowNested=*/true);
1433    }
1434
1435    // Repeatedly split the edge as necessary.
1436    // This is important for nested logical expressions (||, &&, ?:) where we
1437    // want to show all the levels of context.
1438    while (true) {
1439      const Stmt *Dst = Piece->getEndLocation().getStmtOrNull();
1440
1441      // We are looking at an edge. Is the destination within a larger
1442      // expression?
1443      PathDiagnosticLocation DstContext =
1444          getEnclosingStmtLocation(Dst, LC, /*allowNested=*/true);
1445      if (!DstContext.isValid() || DstContext.asStmt() == Dst)
1446        break;
1447
1448      // If the source is in the same context, we're already good.
1449      if (llvm::find(SrcContexts, DstContext) != SrcContexts.end())
1450        break;
1451
1452      // Update the subexpression node to point to the context edge.
1453      Piece->setStartLocation(DstContext);
1454
1455      // Try to extend the previous edge if it's at the same level as the source
1456      // context.
1457      if (Prev != E) {
1458        auto *PrevPiece = dyn_cast<PathDiagnosticControlFlowPiece>(Prev->get());
1459
1460        if (PrevPiece) {
1461          if (const Stmt *PrevSrc =
1462                  PrevPiece->getStartLocation().getStmtOrNull()) {
1463            const Stmt *PrevSrcParent = getStmtParent(PrevSrc, PM);
1464            if (PrevSrcParent ==
1465                getStmtParent(DstContext.getStmtOrNull(), PM)) {
1466              PrevPiece->setEndLocation(DstContext);
1467              break;
1468            }
1469          }
1470        }
1471      }
1472
1473      // Otherwise, split the current edge into a context edge and a
1474      // subexpression edge. Note that the context statement may itself have
1475      // context.
1476      auto P =
1477          std::make_shared<PathDiagnosticControlFlowPiece>(SrcLoc, DstContext);
1478      Piece = P.get();
1479      I = pieces.insert(I, std::move(P));
1480    }
1481  }
1482}
1483
1484/// Move edges from a branch condition to a branch target
1485///        when the condition is simple.
1486///
1487/// This restructures some of the work of addContextEdges.  That function
1488/// creates edges this may destroy, but they work together to create a more
1489/// aesthetically set of edges around branches.  After the call to
1490/// addContextEdges, we may have (1) an edge to the branch, (2) an edge from
1491/// the branch to the branch condition, and (3) an edge from the branch
1492/// condition to the branch target.  We keep (1), but may wish to remove (2)
1493/// and move the source of (3) to the branch if the branch condition is simple.
1494static void simplifySimpleBranches(PathPieces &pieces) {
1495  for (PathPieces::iterator I = pieces.begin(), E = pieces.end(); I != E; ++I) {
1496    const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1497
1498    if (!PieceI)
1499      continue;
1500
1501    const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1502    const Stmt *s1End   = PieceI->getEndLocation().getStmtOrNull();
1503
1504    if (!s1Start || !s1End)
1505      continue;
1506
1507    PathPieces::iterator NextI = I; ++NextI;
1508    if (NextI == E)
1509      break;
1510
1511    PathDiagnosticControlFlowPiece *PieceNextI = nullptr;
1512
1513    while (true) {
1514      if (NextI == E)
1515        break;
1516
1517      const auto *EV = dyn_cast<PathDiagnosticEventPiece>(NextI->get());
1518      if (EV) {
1519        StringRef S = EV->getString();
1520        if (S == StrEnteringLoop || S == StrLoopBodyZero ||
1521            S == StrLoopCollectionEmpty || S == StrLoopRangeEmpty) {
1522          ++NextI;
1523          continue;
1524        }
1525        break;
1526      }
1527
1528      PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1529      break;
1530    }
1531
1532    if (!PieceNextI)
1533      continue;
1534
1535    const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1536    const Stmt *s2End   = PieceNextI->getEndLocation().getStmtOrNull();
1537
1538    if (!s2Start || !s2End || s1End != s2Start)
1539      continue;
1540
1541    // We only perform this transformation for specific branch kinds.
1542    // We don't want to do this for do..while, for example.
1543    if (!(isa<ForStmt>(s1Start) || isa<WhileStmt>(s1Start) ||
1544          isa<IfStmt>(s1Start) || isa<ObjCForCollectionStmt>(s1Start) ||
1545          isa<CXXForRangeStmt>(s1Start)))
1546      continue;
1547
1548    // Is s1End the branch condition?
1549    if (!isConditionForTerminator(s1Start, s1End))
1550      continue;
1551
1552    // Perform the hoisting by eliminating (2) and changing the start
1553    // location of (3).
1554    PieceNextI->setStartLocation(PieceI->getStartLocation());
1555    I = pieces.erase(I);
1556  }
1557}
1558
1559/// Returns the number of bytes in the given (character-based) SourceRange.
1560///
1561/// If the locations in the range are not on the same line, returns None.
1562///
1563/// Note that this does not do a precise user-visible character or column count.
1564static Optional<size_t> getLengthOnSingleLine(const SourceManager &SM,
1565                                              SourceRange Range) {
1566  SourceRange ExpansionRange(SM.getExpansionLoc(Range.getBegin()),
1567                             SM.getExpansionRange(Range.getEnd()).getEnd());
1568
1569  FileID FID = SM.getFileID(ExpansionRange.getBegin());
1570  if (FID != SM.getFileID(ExpansionRange.getEnd()))
1571    return None;
1572
1573  bool Invalid;
1574  const llvm::MemoryBuffer *Buffer = SM.getBuffer(FID, &Invalid);
1575  if (Invalid)
1576    return None;
1577
1578  unsigned BeginOffset = SM.getFileOffset(ExpansionRange.getBegin());
1579  unsigned EndOffset = SM.getFileOffset(ExpansionRange.getEnd());
1580  StringRef Snippet = Buffer->getBuffer().slice(BeginOffset, EndOffset);
1581
1582  // We're searching the raw bytes of the buffer here, which might include
1583  // escaped newlines and such. That's okay; we're trying to decide whether the
1584  // SourceRange is covering a large or small amount of space in the user's
1585  // editor.
1586  if (Snippet.find_first_of("\r\n") != StringRef::npos)
1587    return None;
1588
1589  // This isn't Unicode-aware, but it doesn't need to be.
1590  return Snippet.size();
1591}
1592
1593/// \sa getLengthOnSingleLine(SourceManager, SourceRange)
1594static Optional<size_t> getLengthOnSingleLine(const SourceManager &SM,
1595                                              const Stmt *S) {
1596  return getLengthOnSingleLine(SM, S->getSourceRange());
1597}
1598
1599/// Eliminate two-edge cycles created by addContextEdges().
1600///
1601/// Once all the context edges are in place, there are plenty of cases where
1602/// there's a single edge from a top-level statement to a subexpression,
1603/// followed by a single path note, and then a reverse edge to get back out to
1604/// the top level. If the statement is simple enough, the subexpression edges
1605/// just add noise and make it harder to understand what's going on.
1606///
1607/// This function only removes edges in pairs, because removing only one edge
1608/// might leave other edges dangling.
1609///
1610/// This will not remove edges in more complicated situations:
1611/// - if there is more than one "hop" leading to or from a subexpression.
1612/// - if there is an inlined call between the edges instead of a single event.
1613/// - if the whole statement is large enough that having subexpression arrows
1614///   might be helpful.
1615static void removeContextCycles(PathPieces &Path, const SourceManager &SM) {
1616  for (PathPieces::iterator I = Path.begin(), E = Path.end(); I != E; ) {
1617    // Pattern match the current piece and its successor.
1618    const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1619
1620    if (!PieceI) {
1621      ++I;
1622      continue;
1623    }
1624
1625    const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1626    const Stmt *s1End   = PieceI->getEndLocation().getStmtOrNull();
1627
1628    PathPieces::iterator NextI = I; ++NextI;
1629    if (NextI == E)
1630      break;
1631
1632    const auto *PieceNextI =
1633        dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1634
1635    if (!PieceNextI) {
1636      if (isa<PathDiagnosticEventPiece>(NextI->get())) {
1637        ++NextI;
1638        if (NextI == E)
1639          break;
1640        PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1641      }
1642
1643      if (!PieceNextI) {
1644        ++I;
1645        continue;
1646      }
1647    }
1648
1649    const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1650    const Stmt *s2End   = PieceNextI->getEndLocation().getStmtOrNull();
1651
1652    if (s1Start && s2Start && s1Start == s2End && s2Start == s1End) {
1653      const size_t MAX_SHORT_LINE_LENGTH = 80;
1654      Optional<size_t> s1Length = getLengthOnSingleLine(SM, s1Start);
1655      if (s1Length && *s1Length <= MAX_SHORT_LINE_LENGTH) {
1656        Optional<size_t> s2Length = getLengthOnSingleLine(SM, s2Start);
1657        if (s2Length && *s2Length <= MAX_SHORT_LINE_LENGTH) {
1658          Path.erase(I);
1659          I = Path.erase(NextI);
1660          continue;
1661        }
1662      }
1663    }
1664
1665    ++I;
1666  }
1667}
1668
1669/// Return true if X is contained by Y.
1670static bool lexicalContains(const ParentMap &PM, const Stmt *X, const Stmt *Y) {
1671  while (X) {
1672    if (X == Y)
1673      return true;
1674    X = PM.getParent(X);
1675  }
1676  return false;
1677}
1678
1679// Remove short edges on the same line less than 3 columns in difference.
1680static void removePunyEdges(PathPieces &path, const SourceManager &SM,
1681                            const ParentMap &PM) {
1682  bool erased = false;
1683
1684  for (PathPieces::iterator I = path.begin(), E = path.end(); I != E;
1685       erased ? I : ++I) {
1686    erased = false;
1687
1688    const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1689
1690    if (!PieceI)
1691      continue;
1692
1693    const Stmt *start = PieceI->getStartLocation().getStmtOrNull();
1694    const Stmt *end   = PieceI->getEndLocation().getStmtOrNull();
1695
1696    if (!start || !end)
1697      continue;
1698
1699    const Stmt *endParent = PM.getParent(end);
1700    if (!endParent)
1701      continue;
1702
1703    if (isConditionForTerminator(end, endParent))
1704      continue;
1705
1706    SourceLocation FirstLoc = start->getBeginLoc();
1707    SourceLocation SecondLoc = end->getBeginLoc();
1708
1709    if (!SM.isWrittenInSameFile(FirstLoc, SecondLoc))
1710      continue;
1711    if (SM.isBeforeInTranslationUnit(SecondLoc, FirstLoc))
1712      std::swap(SecondLoc, FirstLoc);
1713
1714    SourceRange EdgeRange(FirstLoc, SecondLoc);
1715    Optional<size_t> ByteWidth = getLengthOnSingleLine(SM, EdgeRange);
1716
1717    // If the statements are on different lines, continue.
1718    if (!ByteWidth)
1719      continue;
1720
1721    const size_t MAX_PUNY_EDGE_LENGTH = 2;
1722    if (*ByteWidth <= MAX_PUNY_EDGE_LENGTH) {
1723      // FIXME: There are enough /bytes/ between the endpoints of the edge, but
1724      // there might not be enough /columns/. A proper user-visible column count
1725      // is probably too expensive, though.
1726      I = path.erase(I);
1727      erased = true;
1728      continue;
1729    }
1730  }
1731}
1732
1733static void removeIdenticalEvents(PathPieces &path) {
1734  for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ++I) {
1735    const auto *PieceI = dyn_cast<PathDiagnosticEventPiece>(I->get());
1736
1737    if (!PieceI)
1738      continue;
1739
1740    PathPieces::iterator NextI = I; ++NextI;
1741    if (NextI == E)
1742      return;
1743
1744    const auto *PieceNextI = dyn_cast<PathDiagnosticEventPiece>(NextI->get());
1745
1746    if (!PieceNextI)
1747      continue;
1748
1749    // Erase the second piece if it has the same exact message text.
1750    if (PieceI->getString() == PieceNextI->getString()) {
1751      path.erase(NextI);
1752    }
1753  }
1754}
1755
1756static bool optimizeEdges(const PathDiagnosticConstruct &C, PathPieces &path,
1757                          OptimizedCallsSet &OCS) {
1758  bool hasChanges = false;
1759  const LocationContext *LC = C.getLocationContextFor(&path);
1760  assert(LC);
1761  const ParentMap &PM = LC->getParentMap();
1762  const SourceManager &SM = C.getSourceManager();
1763
1764  for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ) {
1765    // Optimize subpaths.
1766    if (auto *CallI = dyn_cast<PathDiagnosticCallPiece>(I->get())) {
1767      // Record the fact that a call has been optimized so we only do the
1768      // effort once.
1769      if (!OCS.count(CallI)) {
1770        while (optimizeEdges(C, CallI->path, OCS)) {
1771        }
1772        OCS.insert(CallI);
1773      }
1774      ++I;
1775      continue;
1776    }
1777
1778    // Pattern match the current piece and its successor.
1779    auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1780
1781    if (!PieceI) {
1782      ++I;
1783      continue;
1784    }
1785
1786    const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1787    const Stmt *s1End   = PieceI->getEndLocation().getStmtOrNull();
1788    const Stmt *level1 = getStmtParent(s1Start, PM);
1789    const Stmt *level2 = getStmtParent(s1End, PM);
1790
1791    PathPieces::iterator NextI = I; ++NextI;
1792    if (NextI == E)
1793      break;
1794
1795    const auto *PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1796
1797    if (!PieceNextI) {
1798      ++I;
1799      continue;
1800    }
1801
1802    const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1803    const Stmt *s2End   = PieceNextI->getEndLocation().getStmtOrNull();
1804    const Stmt *level3 = getStmtParent(s2Start, PM);
1805    const Stmt *level4 = getStmtParent(s2End, PM);
1806
1807    // Rule I.
1808    //
1809    // If we have two consecutive control edges whose end/begin locations
1810    // are at the same level (e.g. statements or top-level expressions within
1811    // a compound statement, or siblings share a single ancestor expression),
1812    // then merge them if they have no interesting intermediate event.
1813    //
1814    // For example:
1815    //
1816    // (1.1 -> 1.2) -> (1.2 -> 1.3) becomes (1.1 -> 1.3) because the common
1817    // parent is '1'.  Here 'x.y.z' represents the hierarchy of statements.
1818    //
1819    // NOTE: this will be limited later in cases where we add barriers
1820    // to prevent this optimization.
1821    if (level1 && level1 == level2 && level1 == level3 && level1 == level4) {
1822      PieceI->setEndLocation(PieceNextI->getEndLocation());
1823      path.erase(NextI);
1824      hasChanges = true;
1825      continue;
1826    }
1827
1828    // Rule II.
1829    //
1830    // Eliminate edges between subexpressions and parent expressions
1831    // when the subexpression is consumed.
1832    //
1833    // NOTE: this will be limited later in cases where we add barriers
1834    // to prevent this optimization.
1835    if (s1End && s1End == s2Start && level2) {
1836      bool removeEdge = false;
1837      // Remove edges into the increment or initialization of a
1838      // loop that have no interleaving event.  This means that
1839      // they aren't interesting.
1840      if (isIncrementOrInitInForLoop(s1End, level2))
1841        removeEdge = true;
1842      // Next only consider edges that are not anchored on
1843      // the condition of a terminator.  This are intermediate edges
1844      // that we might want to trim.
1845      else if (!isConditionForTerminator(level2, s1End)) {
1846        // Trim edges on expressions that are consumed by
1847        // the parent expression.
1848        if (isa<Expr>(s1End) && PM.isConsumedExpr(cast<Expr>(s1End))) {
1849          removeEdge = true;
1850        }
1851        // Trim edges where a lexical containment doesn't exist.
1852        // For example:
1853        //
1854        //  X -> Y -> Z
1855        //
1856        // If 'Z' lexically contains Y (it is an ancestor) and
1857        // 'X' does not lexically contain Y (it is a descendant OR
1858        // it has no lexical relationship at all) then trim.
1859        //
1860        // This can eliminate edges where we dive into a subexpression
1861        // and then pop back out, etc.
1862        else if (s1Start && s2End &&
1863                 lexicalContains(PM, s2Start, s2End) &&
1864                 !lexicalContains(PM, s1End, s1Start)) {
1865          removeEdge = true;
1866        }
1867        // Trim edges from a subexpression back to the top level if the
1868        // subexpression is on a different line.
1869        //
1870        // A.1 -> A -> B
1871        // becomes
1872        // A.1 -> B
1873        //
1874        // These edges just look ugly and don't usually add anything.
1875        else if (s1Start && s2End &&
1876                 lexicalContains(PM, s1Start, s1End)) {
1877          SourceRange EdgeRange(PieceI->getEndLocation().asLocation(),
1878                                PieceI->getStartLocation().asLocation());
1879          if (!getLengthOnSingleLine(SM, EdgeRange).hasValue())
1880            removeEdge = true;
1881        }
1882      }
1883
1884      if (removeEdge) {
1885        PieceI->setEndLocation(PieceNextI->getEndLocation());
1886        path.erase(NextI);
1887        hasChanges = true;
1888        continue;
1889      }
1890    }
1891
1892    // Optimize edges for ObjC fast-enumeration loops.
1893    //
1894    // (X -> collection) -> (collection -> element)
1895    //
1896    // becomes:
1897    //
1898    // (X -> element)
1899    if (s1End == s2Start) {
1900      const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(level3);
1901      if (FS && FS->getCollection()->IgnoreParens() == s2Start &&
1902          s2End == FS->getElement()) {
1903        PieceI->setEndLocation(PieceNextI->getEndLocation());
1904        path.erase(NextI);
1905        hasChanges = true;
1906        continue;
1907      }
1908    }
1909
1910    // No changes at this index?  Move to the next one.
1911    ++I;
1912  }
1913
1914  if (!hasChanges) {
1915    // Adjust edges into subexpressions to make them more uniform
1916    // and aesthetically pleasing.
1917    addContextEdges(path, LC);
1918    // Remove "cyclical" edges that include one or more context edges.
1919    removeContextCycles(path, SM);
1920    // Hoist edges originating from branch conditions to branches
1921    // for simple branches.
1922    simplifySimpleBranches(path);
1923    // Remove any puny edges left over after primary optimization pass.
1924    removePunyEdges(path, SM, PM);
1925    // Remove identical events.
1926    removeIdenticalEvents(path);
1927  }
1928
1929  return hasChanges;
1930}
1931
1932/// Drop the very first edge in a path, which should be a function entry edge.
1933///
1934/// If the first edge is not a function entry edge (say, because the first
1935/// statement had an invalid source location), this function does nothing.
1936// FIXME: We should just generate invalid edges anyway and have the optimizer
1937// deal with them.
1938static void dropFunctionEntryEdge(const PathDiagnosticConstruct &C,
1939                                  PathPieces &Path) {
1940  const auto *FirstEdge =
1941      dyn_cast<PathDiagnosticControlFlowPiece>(Path.front().get());
1942  if (!FirstEdge)
1943    return;
1944
1945  const Decl *D = C.getLocationContextFor(&Path)->getDecl();
1946  PathDiagnosticLocation EntryLoc =
1947      PathDiagnosticLocation::createBegin(D, C.getSourceManager());
1948  if (FirstEdge->getStartLocation() != EntryLoc)
1949    return;
1950
1951  Path.pop_front();
1952}
1953
1954/// Populate executes lines with lines containing at least one diagnostics.
1955static void updateExecutedLinesWithDiagnosticPieces(PathDiagnostic &PD) {
1956
1957  PathPieces path = PD.path.flatten(/*ShouldFlattenMacros=*/true);
1958  FilesToLineNumsMap &ExecutedLines = PD.getExecutedLines();
1959
1960  for (const auto &P : path) {
1961    FullSourceLoc Loc = P->getLocation().asLocation().getExpansionLoc();
1962    FileID FID = Loc.getFileID();
1963    unsigned LineNo = Loc.getLineNumber();
1964    assert(FID.isValid());
1965    ExecutedLines[FID].insert(LineNo);
1966  }
1967}
1968
1969PathDiagnosticConstruct::PathDiagnosticConstruct(
1970    const PathDiagnosticConsumer *PDC, const ExplodedNode *ErrorNode,
1971    const PathSensitiveBugReport *R)
1972    : Consumer(PDC), CurrentNode(ErrorNode),
1973      SM(CurrentNode->getCodeDecl().getASTContext().getSourceManager()),
1974      PD(generateEmptyDiagnosticForReport(R, getSourceManager())) {
1975  LCM[&PD->getActivePath()] = ErrorNode->getLocationContext();
1976}
1977
1978PathDiagnosticBuilder::PathDiagnosticBuilder(
1979    BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath,
1980    PathSensitiveBugReport *r, const ExplodedNode *ErrorNode,
1981    std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics)
1982    : BugReporterContext(BRC), BugPath(std::move(BugPath)), R(r),
1983      ErrorNode(ErrorNode),
1984      VisitorsDiagnostics(std::move(VisitorsDiagnostics)) {}
1985
1986std::unique_ptr<PathDiagnostic>
1987PathDiagnosticBuilder::generate(const PathDiagnosticConsumer *PDC) const {
1988  PathDiagnosticConstruct Construct(PDC, ErrorNode, R);
1989
1990  const SourceManager &SM = getSourceManager();
1991  const AnalyzerOptions &Opts = getAnalyzerOptions();
1992  StringRef ErrorTag = ErrorNode->getLocation().getTag()->getTagDescription();
1993
1994  // See whether we need to silence the checker/package.
1995  // FIXME: This will not work if the report was emitted with an incorrect tag.
1996  for (const std::string &CheckerOrPackage : Opts.SilencedCheckersAndPackages) {
1997    if (ErrorTag.startswith(CheckerOrPackage))
1998      return nullptr;
1999  }
2000
2001  if (!PDC->shouldGenerateDiagnostics())
2002    return generateEmptyDiagnosticForReport(R, getSourceManager());
2003
2004  // Construct the final (warning) event for the bug report.
2005  auto EndNotes = VisitorsDiagnostics->find(ErrorNode);
2006  PathDiagnosticPieceRef LastPiece;
2007  if (EndNotes != VisitorsDiagnostics->end()) {
2008    assert(!EndNotes->second.empty());
2009    LastPiece = EndNotes->second[0];
2010  } else {
2011    LastPiece = BugReporterVisitor::getDefaultEndPath(*this, ErrorNode,
2012                                                      *getBugReport());
2013  }
2014  Construct.PD->setEndOfPath(LastPiece);
2015
2016  PathDiagnosticLocation PrevLoc = Construct.PD->getLocation();
2017  // From the error node to the root, ascend the bug path and construct the bug
2018  // report.
2019  while (Construct.ascendToPrevNode()) {
2020    generatePathDiagnosticsForNode(Construct, PrevLoc);
2021
2022    auto VisitorNotes = VisitorsDiagnostics->find(Construct.getCurrentNode());
2023    if (VisitorNotes == VisitorsDiagnostics->end())
2024      continue;
2025
2026    // This is a workaround due to inability to put shared PathDiagnosticPiece
2027    // into a FoldingSet.
2028    std::set<llvm::FoldingSetNodeID> DeduplicationSet;
2029
2030    // Add pieces from custom visitors.
2031    for (const PathDiagnosticPieceRef &Note : VisitorNotes->second) {
2032      llvm::FoldingSetNodeID ID;
2033      Note->Profile(ID);
2034      if (!DeduplicationSet.insert(ID).second)
2035        continue;
2036
2037      if (PDC->shouldAddPathEdges())
2038        addEdgeToPath(Construct.getActivePath(), PrevLoc, Note->getLocation());
2039      updateStackPiecesWithMessage(Note, Construct.CallStack);
2040      Construct.getActivePath().push_front(Note);
2041    }
2042  }
2043
2044  if (PDC->shouldAddPathEdges()) {
2045    // Add an edge to the start of the function.
2046    // We'll prune it out later, but it helps make diagnostics more uniform.
2047    const StackFrameContext *CalleeLC =
2048        Construct.getLocationContextForActivePath()->getStackFrame();
2049    const Decl *D = CalleeLC->getDecl();
2050    addEdgeToPath(Construct.getActivePath(), PrevLoc,
2051                  PathDiagnosticLocation::createBegin(D, SM));
2052  }
2053
2054
2055  // Finally, prune the diagnostic path of uninteresting stuff.
2056  if (!Construct.PD->path.empty()) {
2057    if (R->shouldPrunePath() && Opts.ShouldPrunePaths) {
2058      bool stillHasNotes =
2059          removeUnneededCalls(Construct, Construct.getMutablePieces(), R);
2060      assert(stillHasNotes);
2061      (void)stillHasNotes;
2062    }
2063
2064    // Remove pop-up notes if needed.
2065    if (!Opts.ShouldAddPopUpNotes)
2066      removePopUpNotes(Construct.getMutablePieces());
2067
2068    // Redirect all call pieces to have valid locations.
2069    adjustCallLocations(Construct.getMutablePieces());
2070    removePiecesWithInvalidLocations(Construct.getMutablePieces());
2071
2072    if (PDC->shouldAddPathEdges()) {
2073
2074      // Reduce the number of edges from a very conservative set
2075      // to an aesthetically pleasing subset that conveys the
2076      // necessary information.
2077      OptimizedCallsSet OCS;
2078      while (optimizeEdges(Construct, Construct.getMutablePieces(), OCS)) {
2079      }
2080
2081      // Drop the very first function-entry edge. It's not really necessary
2082      // for top-level functions.
2083      dropFunctionEntryEdge(Construct, Construct.getMutablePieces());
2084    }
2085
2086    // Remove messages that are basically the same, and edges that may not
2087    // make sense.
2088    // We have to do this after edge optimization in the Extensive mode.
2089    removeRedundantMsgs(Construct.getMutablePieces());
2090    removeEdgesToDefaultInitializers(Construct.getMutablePieces());
2091  }
2092
2093  if (Opts.ShouldDisplayMacroExpansions)
2094    CompactMacroExpandedPieces(Construct.getMutablePieces(), SM);
2095
2096  return std::move(Construct.PD);
2097}
2098
2099//===----------------------------------------------------------------------===//
2100// Methods for BugType and subclasses.
2101//===----------------------------------------------------------------------===//
2102
2103void BugType::anchor() {}
2104
2105void BuiltinBug::anchor() {}
2106
2107//===----------------------------------------------------------------------===//
2108// Methods for BugReport and subclasses.
2109//===----------------------------------------------------------------------===//
2110
2111LLVM_ATTRIBUTE_USED static bool
2112isDependency(const CheckerRegistryData &Registry, StringRef CheckerName) {
2113  for (const std::pair<StringRef, StringRef> &Pair : Registry.Dependencies) {
2114    if (Pair.second == CheckerName)
2115      return true;
2116  }
2117  return false;
2118}
2119
2120LLVM_ATTRIBUTE_USED static bool isHidden(const CheckerRegistryData &Registry,
2121                                         StringRef CheckerName) {
2122  for (const CheckerInfo &Checker : Registry.Checkers) {
2123    if (Checker.FullName == CheckerName)
2124      return Checker.IsHidden;
2125  }
2126  llvm_unreachable(
2127      "Checker name not found in CheckerRegistry -- did you retrieve it "
2128      "correctly from CheckerManager::getCurrentCheckerName?");
2129}
2130
2131PathSensitiveBugReport::PathSensitiveBugReport(
2132    const BugType &bt, StringRef shortDesc, StringRef desc,
2133    const ExplodedNode *errorNode, PathDiagnosticLocation LocationToUnique,
2134    const Decl *DeclToUnique)
2135    : BugReport(Kind::PathSensitive, bt, shortDesc, desc), ErrorNode(errorNode),
2136      ErrorNodeRange(getStmt() ? getStmt()->getSourceRange() : SourceRange()),
2137      UniqueingLocation(LocationToUnique), UniqueingDecl(DeclToUnique) {
2138  assert(!isDependency(ErrorNode->getState()
2139                           ->getAnalysisManager()
2140                           .getCheckerManager()
2141                           ->getCheckerRegistryData(),
2142                       bt.getCheckerName()) &&
2143         "Some checkers depend on this one! We don't allow dependency "
2144         "checkers to emit warnings, because checkers should depend on "
2145         "*modeling*, not *diagnostics*.");
2146
2147  assert(
2148      (bt.getCheckerName().startswith("debug") ||
2149       !isHidden(ErrorNode->getState()
2150                     ->getAnalysisManager()
2151                     .getCheckerManager()
2152                     ->getCheckerRegistryData(),
2153                 bt.getCheckerName())) &&
2154          "Hidden checkers musn't emit diagnostics as they are by definition "
2155          "non-user facing!");
2156}
2157
2158void PathSensitiveBugReport::addVisitor(
2159    std::unique_ptr<BugReporterVisitor> visitor) {
2160  if (!visitor)
2161    return;
2162
2163  llvm::FoldingSetNodeID ID;
2164  visitor->Profile(ID);
2165
2166  void *InsertPos = nullptr;
2167  if (CallbacksSet.FindNodeOrInsertPos(ID, InsertPos)) {
2168    return;
2169  }
2170
2171  Callbacks.push_back(std::move(visitor));
2172}
2173
2174void PathSensitiveBugReport::clearVisitors() {
2175  Callbacks.clear();
2176}
2177
2178const Decl *PathSensitiveBugReport::getDeclWithIssue() const {
2179  const ExplodedNode *N = getErrorNode();
2180  if (!N)
2181    return nullptr;
2182
2183  const LocationContext *LC = N->getLocationContext();
2184  return LC->getStackFrame()->getDecl();
2185}
2186
2187void BasicBugReport::Profile(llvm::FoldingSetNodeID& hash) const {
2188  hash.AddInteger(static_cast<int>(getKind()));
2189  hash.AddPointer(&BT);
2190  hash.AddString(Description);
2191  assert(Location.isValid());
2192  Location.Profile(hash);
2193
2194  for (SourceRange range : Ranges) {
2195    if (!range.isValid())
2196      continue;
2197    hash.AddInteger(range.getBegin().getRawEncoding());
2198    hash.AddInteger(range.getEnd().getRawEncoding());
2199  }
2200}
2201
2202void PathSensitiveBugReport::Profile(llvm::FoldingSetNodeID &hash) const {
2203  hash.AddInteger(static_cast<int>(getKind()));
2204  hash.AddPointer(&BT);
2205  hash.AddString(Description);
2206  PathDiagnosticLocation UL = getUniqueingLocation();
2207  if (UL.isValid()) {
2208    UL.Profile(hash);
2209  } else {
2210    // TODO: The statement may be null if the report was emitted before any
2211    // statements were executed. In particular, some checkers by design
2212    // occasionally emit their reports in empty functions (that have no
2213    // statements in their body). Do we profile correctly in this case?
2214    hash.AddPointer(ErrorNode->getCurrentOrPreviousStmtForDiagnostics());
2215  }
2216
2217  for (SourceRange range : Ranges) {
2218    if (!range.isValid())
2219      continue;
2220    hash.AddInteger(range.getBegin().getRawEncoding());
2221    hash.AddInteger(range.getEnd().getRawEncoding());
2222  }
2223}
2224
2225template <class T>
2226static void insertToInterestingnessMap(
2227    llvm::DenseMap<T, bugreporter::TrackingKind> &InterestingnessMap, T Val,
2228    bugreporter::TrackingKind TKind) {
2229  auto Result = InterestingnessMap.insert({Val, TKind});
2230
2231  if (Result.second)
2232    return;
2233
2234  // Even if this symbol/region was already marked as interesting as a
2235  // condition, if we later mark it as interesting again but with
2236  // thorough tracking, overwrite it. Entities marked with thorough
2237  // interestiness are the most important (or most interesting, if you will),
2238  // and we wouldn't like to downplay their importance.
2239
2240  switch (TKind) {
2241    case bugreporter::TrackingKind::Thorough:
2242      Result.first->getSecond() = bugreporter::TrackingKind::Thorough;
2243      return;
2244    case bugreporter::TrackingKind::Condition:
2245      return;
2246    }
2247
2248    llvm_unreachable(
2249        "BugReport::markInteresting currently can only handle 2 different "
2250        "tracking kinds! Please define what tracking kind should this entitiy"
2251        "have, if it was already marked as interesting with a different kind!");
2252}
2253
2254void PathSensitiveBugReport::markInteresting(SymbolRef sym,
2255                                             bugreporter::TrackingKind TKind) {
2256  if (!sym)
2257    return;
2258
2259  insertToInterestingnessMap(InterestingSymbols, sym, TKind);
2260
2261  if (const auto *meta = dyn_cast<SymbolMetadata>(sym))
2262    markInteresting(meta->getRegion(), TKind);
2263}
2264
2265void PathSensitiveBugReport::markInteresting(const MemRegion *R,
2266                                             bugreporter::TrackingKind TKind) {
2267  if (!R)
2268    return;
2269
2270  R = R->getBaseRegion();
2271  insertToInterestingnessMap(InterestingRegions, R, TKind);
2272
2273  if (const auto *SR = dyn_cast<SymbolicRegion>(R))
2274    markInteresting(SR->getSymbol(), TKind);
2275}
2276
2277void PathSensitiveBugReport::markInteresting(SVal V,
2278                                             bugreporter::TrackingKind TKind) {
2279  markInteresting(V.getAsRegion(), TKind);
2280  markInteresting(V.getAsSymbol(), TKind);
2281}
2282
2283void PathSensitiveBugReport::markInteresting(const LocationContext *LC) {
2284  if (!LC)
2285    return;
2286  InterestingLocationContexts.insert(LC);
2287}
2288
2289Optional<bugreporter::TrackingKind>
2290PathSensitiveBugReport::getInterestingnessKind(SVal V) const {
2291  auto RKind = getInterestingnessKind(V.getAsRegion());
2292  auto SKind = getInterestingnessKind(V.getAsSymbol());
2293  if (!RKind)
2294    return SKind;
2295  if (!SKind)
2296    return RKind;
2297
2298  // If either is marked with throrough tracking, return that, we wouldn't like
2299  // to downplay a note's importance by 'only' mentioning it as a condition.
2300  switch(*RKind) {
2301    case bugreporter::TrackingKind::Thorough:
2302      return RKind;
2303    case bugreporter::TrackingKind::Condition:
2304      return SKind;
2305  }
2306
2307  llvm_unreachable(
2308      "BugReport::getInterestingnessKind currently can only handle 2 different "
2309      "tracking kinds! Please define what tracking kind should we return here "
2310      "when the kind of getAsRegion() and getAsSymbol() is different!");
2311  return None;
2312}
2313
2314Optional<bugreporter::TrackingKind>
2315PathSensitiveBugReport::getInterestingnessKind(SymbolRef sym) const {
2316  if (!sym)
2317    return None;
2318  // We don't currently consider metadata symbols to be interesting
2319  // even if we know their region is interesting. Is that correct behavior?
2320  auto It = InterestingSymbols.find(sym);
2321  if (It == InterestingSymbols.end())
2322    return None;
2323  return It->getSecond();
2324}
2325
2326Optional<bugreporter::TrackingKind>
2327PathSensitiveBugReport::getInterestingnessKind(const MemRegion *R) const {
2328  if (!R)
2329    return None;
2330
2331  R = R->getBaseRegion();
2332  auto It = InterestingRegions.find(R);
2333  if (It != InterestingRegions.end())
2334    return It->getSecond();
2335
2336  if (const auto *SR = dyn_cast<SymbolicRegion>(R))
2337    return getInterestingnessKind(SR->getSymbol());
2338  return None;
2339}
2340
2341bool PathSensitiveBugReport::isInteresting(SVal V) const {
2342  return getInterestingnessKind(V).hasValue();
2343}
2344
2345bool PathSensitiveBugReport::isInteresting(SymbolRef sym) const {
2346  return getInterestingnessKind(sym).hasValue();
2347}
2348
2349bool PathSensitiveBugReport::isInteresting(const MemRegion *R) const {
2350  return getInterestingnessKind(R).hasValue();
2351}
2352
2353bool PathSensitiveBugReport::isInteresting(const LocationContext *LC)  const {
2354  if (!LC)
2355    return false;
2356  return InterestingLocationContexts.count(LC);
2357}
2358
2359const Stmt *PathSensitiveBugReport::getStmt() const {
2360  if (!ErrorNode)
2361    return nullptr;
2362
2363  ProgramPoint ProgP = ErrorNode->getLocation();
2364  const Stmt *S = nullptr;
2365
2366  if (Optional<BlockEntrance> BE = ProgP.getAs<BlockEntrance>()) {
2367    CFGBlock &Exit = ProgP.getLocationContext()->getCFG()->getExit();
2368    if (BE->getBlock() == &Exit)
2369      S = ErrorNode->getPreviousStmtForDiagnostics();
2370  }
2371  if (!S)
2372    S = ErrorNode->getStmtForDiagnostics();
2373
2374  return S;
2375}
2376
2377ArrayRef<SourceRange>
2378PathSensitiveBugReport::getRanges() const {
2379  // If no custom ranges, add the range of the statement corresponding to
2380  // the error node.
2381  if (Ranges.empty() && isa_and_nonnull<Expr>(getStmt()))
2382      return ErrorNodeRange;
2383
2384  return Ranges;
2385}
2386
2387PathDiagnosticLocation
2388PathSensitiveBugReport::getLocation() const {
2389  assert(ErrorNode && "Cannot create a location with a null node.");
2390  const Stmt *S = ErrorNode->getStmtForDiagnostics();
2391    ProgramPoint P = ErrorNode->getLocation();
2392  const LocationContext *LC = P.getLocationContext();
2393  SourceManager &SM =
2394      ErrorNode->getState()->getStateManager().getContext().getSourceManager();
2395
2396  if (!S) {
2397    // If this is an implicit call, return the implicit call point location.
2398    if (Optional<PreImplicitCall> PIE = P.getAs<PreImplicitCall>())
2399      return PathDiagnosticLocation(PIE->getLocation(), SM);
2400    if (auto FE = P.getAs<FunctionExitPoint>()) {
2401      if (const ReturnStmt *RS = FE->getStmt())
2402        return PathDiagnosticLocation::createBegin(RS, SM, LC);
2403    }
2404    S = ErrorNode->getNextStmtForDiagnostics();
2405  }
2406
2407  if (S) {
2408    // For member expressions, return the location of the '.' or '->'.
2409    if (const auto *ME = dyn_cast<MemberExpr>(S))
2410      return PathDiagnosticLocation::createMemberLoc(ME, SM);
2411
2412    // For binary operators, return the location of the operator.
2413    if (const auto *B = dyn_cast<BinaryOperator>(S))
2414      return PathDiagnosticLocation::createOperatorLoc(B, SM);
2415
2416    if (P.getAs<PostStmtPurgeDeadSymbols>())
2417      return PathDiagnosticLocation::createEnd(S, SM, LC);
2418
2419    if (S->getBeginLoc().isValid())
2420      return PathDiagnosticLocation(S, SM, LC);
2421
2422    return PathDiagnosticLocation(
2423        PathDiagnosticLocation::getValidSourceLocation(S, LC), SM);
2424  }
2425
2426  return PathDiagnosticLocation::createDeclEnd(ErrorNode->getLocationContext(),
2427                                               SM);
2428}
2429
2430//===----------------------------------------------------------------------===//
2431// Methods for BugReporter and subclasses.
2432//===----------------------------------------------------------------------===//
2433
2434const ExplodedGraph &PathSensitiveBugReporter::getGraph() const {
2435  return Eng.getGraph();
2436}
2437
2438ProgramStateManager &PathSensitiveBugReporter::getStateManager() const {
2439  return Eng.getStateManager();
2440}
2441
2442BugReporter::BugReporter(BugReporterData &d) : D(d) {}
2443BugReporter::~BugReporter() {
2444  // Make sure reports are flushed.
2445  assert(StrBugTypes.empty() &&
2446         "Destroying BugReporter before diagnostics are emitted!");
2447
2448  // Free the bug reports we are tracking.
2449  for (const auto I : EQClassesVector)
2450    delete I;
2451}
2452
2453void BugReporter::FlushReports() {
2454  // We need to flush reports in deterministic order to ensure the order
2455  // of the reports is consistent between runs.
2456  for (const auto EQ : EQClassesVector)
2457    FlushReport(*EQ);
2458
2459  // BugReporter owns and deletes only BugTypes created implicitly through
2460  // EmitBasicReport.
2461  // FIXME: There are leaks from checkers that assume that the BugTypes they
2462  // create will be destroyed by the BugReporter.
2463  StrBugTypes.clear();
2464}
2465
2466//===----------------------------------------------------------------------===//
2467// PathDiagnostics generation.
2468//===----------------------------------------------------------------------===//
2469
2470namespace {
2471
2472/// A wrapper around an ExplodedGraph that contains a single path from the root
2473/// to the error node.
2474class BugPathInfo {
2475public:
2476  std::unique_ptr<ExplodedGraph> BugPath;
2477  PathSensitiveBugReport *Report;
2478  const ExplodedNode *ErrorNode;
2479};
2480
2481/// A wrapper around an ExplodedGraph whose leafs are all error nodes. Can
2482/// conveniently retrieve bug paths from a single error node to the root.
2483class BugPathGetter {
2484  std::unique_ptr<ExplodedGraph> TrimmedGraph;
2485
2486  using PriorityMapTy = llvm::DenseMap<const ExplodedNode *, unsigned>;
2487
2488  /// Assign each node with its distance from the root.
2489  PriorityMapTy PriorityMap;
2490
2491  /// Since the getErrorNode() or BugReport refers to the original ExplodedGraph,
2492  /// we need to pair it to the error node of the constructed trimmed graph.
2493  using ReportNewNodePair =
2494      std::pair<PathSensitiveBugReport *, const ExplodedNode *>;
2495  SmallVector<ReportNewNodePair, 32> ReportNodes;
2496
2497  BugPathInfo CurrentBugPath;
2498
2499  /// A helper class for sorting ExplodedNodes by priority.
2500  template <bool Descending>
2501  class PriorityCompare {
2502    const PriorityMapTy &PriorityMap;
2503
2504  public:
2505    PriorityCompare(const PriorityMapTy &M) : PriorityMap(M) {}
2506
2507    bool operator()(const ExplodedNode *LHS, const ExplodedNode *RHS) const {
2508      PriorityMapTy::const_iterator LI = PriorityMap.find(LHS);
2509      PriorityMapTy::const_iterator RI = PriorityMap.find(RHS);
2510      PriorityMapTy::const_iterator E = PriorityMap.end();
2511
2512      if (LI == E)
2513        return Descending;
2514      if (RI == E)
2515        return !Descending;
2516
2517      return Descending ? LI->second > RI->second
2518                        : LI->second < RI->second;
2519    }
2520
2521    bool operator()(const ReportNewNodePair &LHS,
2522                    const ReportNewNodePair &RHS) const {
2523      return (*this)(LHS.second, RHS.second);
2524    }
2525  };
2526
2527public:
2528  BugPathGetter(const ExplodedGraph *OriginalGraph,
2529                ArrayRef<PathSensitiveBugReport *> &bugReports);
2530
2531  BugPathInfo *getNextBugPath();
2532};
2533
2534} // namespace
2535
2536BugPathGetter::BugPathGetter(const ExplodedGraph *OriginalGraph,
2537                             ArrayRef<PathSensitiveBugReport *> &bugReports) {
2538  SmallVector<const ExplodedNode *, 32> Nodes;
2539  for (const auto I : bugReports) {
2540    assert(I->isValid() &&
2541           "We only allow BugReporterVisitors and BugReporter itself to "
2542           "invalidate reports!");
2543    Nodes.emplace_back(I->getErrorNode());
2544  }
2545
2546  // The trimmed graph is created in the body of the constructor to ensure
2547  // that the DenseMaps have been initialized already.
2548  InterExplodedGraphMap ForwardMap;
2549  TrimmedGraph = OriginalGraph->trim(Nodes, &ForwardMap);
2550
2551  // Find the (first) error node in the trimmed graph.  We just need to consult
2552  // the node map which maps from nodes in the original graph to nodes
2553  // in the new graph.
2554  llvm::SmallPtrSet<const ExplodedNode *, 32> RemainingNodes;
2555
2556  for (PathSensitiveBugReport *Report : bugReports) {
2557    const ExplodedNode *NewNode = ForwardMap.lookup(Report->getErrorNode());
2558    assert(NewNode &&
2559           "Failed to construct a trimmed graph that contains this error "
2560           "node!");
2561    ReportNodes.emplace_back(Report, NewNode);
2562    RemainingNodes.insert(NewNode);
2563  }
2564
2565  assert(!RemainingNodes.empty() && "No error node found in the trimmed graph");
2566
2567  // Perform a forward BFS to find all the shortest paths.
2568  std::queue<const ExplodedNode *> WS;
2569
2570  assert(TrimmedGraph->num_roots() == 1);
2571  WS.push(*TrimmedGraph->roots_begin());
2572  unsigned Priority = 0;
2573
2574  while (!WS.empty()) {
2575    const ExplodedNode *Node = WS.front();
2576    WS.pop();
2577
2578    PriorityMapTy::iterator PriorityEntry;
2579    bool IsNew;
2580    std::tie(PriorityEntry, IsNew) = PriorityMap.insert({Node, Priority});
2581    ++Priority;
2582
2583    if (!IsNew) {
2584      assert(PriorityEntry->second <= Priority);
2585      continue;
2586    }
2587
2588    if (RemainingNodes.erase(Node))
2589      if (RemainingNodes.empty())
2590        break;
2591
2592    for (const ExplodedNode *Succ : Node->succs())
2593      WS.push(Succ);
2594  }
2595
2596  // Sort the error paths from longest to shortest.
2597  llvm::sort(ReportNodes, PriorityCompare<true>(PriorityMap));
2598}
2599
2600BugPathInfo *BugPathGetter::getNextBugPath() {
2601  if (ReportNodes.empty())
2602    return nullptr;
2603
2604  const ExplodedNode *OrigN;
2605  std::tie(CurrentBugPath.Report, OrigN) = ReportNodes.pop_back_val();
2606  assert(PriorityMap.find(OrigN) != PriorityMap.end() &&
2607         "error node not accessible from root");
2608
2609  // Create a new graph with a single path. This is the graph that will be
2610  // returned to the caller.
2611  auto GNew = std::make_unique<ExplodedGraph>();
2612
2613  // Now walk from the error node up the BFS path, always taking the
2614  // predeccessor with the lowest number.
2615  ExplodedNode *Succ = nullptr;
2616  while (true) {
2617    // Create the equivalent node in the new graph with the same state
2618    // and location.
2619    ExplodedNode *NewN = GNew->createUncachedNode(
2620        OrigN->getLocation(), OrigN->getState(),
2621        OrigN->getID(), OrigN->isSink());
2622
2623    // Link up the new node with the previous node.
2624    if (Succ)
2625      Succ->addPredecessor(NewN, *GNew);
2626    else
2627      CurrentBugPath.ErrorNode = NewN;
2628
2629    Succ = NewN;
2630
2631    // Are we at the final node?
2632    if (OrigN->pred_empty()) {
2633      GNew->addRoot(NewN);
2634      break;
2635    }
2636
2637    // Find the next predeccessor node.  We choose the node that is marked
2638    // with the lowest BFS number.
2639    OrigN = *std::min_element(OrigN->pred_begin(), OrigN->pred_end(),
2640                              PriorityCompare<false>(PriorityMap));
2641  }
2642
2643  CurrentBugPath.BugPath = std::move(GNew);
2644
2645  return &CurrentBugPath;
2646}
2647
2648/// CompactMacroExpandedPieces - This function postprocesses a PathDiagnostic
2649/// object and collapses PathDiagosticPieces that are expanded by macros.
2650static void CompactMacroExpandedPieces(PathPieces &path,
2651                                       const SourceManager& SM) {
2652  using MacroStackTy = std::vector<
2653      std::pair<std::shared_ptr<PathDiagnosticMacroPiece>, SourceLocation>>;
2654
2655  using PiecesTy = std::vector<PathDiagnosticPieceRef>;
2656
2657  MacroStackTy MacroStack;
2658  PiecesTy Pieces;
2659
2660  for (PathPieces::const_iterator I = path.begin(), E = path.end();
2661       I != E; ++I) {
2662    const auto &piece = *I;
2663
2664    // Recursively compact calls.
2665    if (auto *call = dyn_cast<PathDiagnosticCallPiece>(&*piece)) {
2666      CompactMacroExpandedPieces(call->path, SM);
2667    }
2668
2669    // Get the location of the PathDiagnosticPiece.
2670    const FullSourceLoc Loc = piece->getLocation().asLocation();
2671
2672    // Determine the instantiation location, which is the location we group
2673    // related PathDiagnosticPieces.
2674    SourceLocation InstantiationLoc = Loc.isMacroID() ?
2675                                      SM.getExpansionLoc(Loc) :
2676                                      SourceLocation();
2677
2678    if (Loc.isFileID()) {
2679      MacroStack.clear();
2680      Pieces.push_back(piece);
2681      continue;
2682    }
2683
2684    assert(Loc.isMacroID());
2685
2686    // Is the PathDiagnosticPiece within the same macro group?
2687    if (!MacroStack.empty() && InstantiationLoc == MacroStack.back().second) {
2688      MacroStack.back().first->subPieces.push_back(piece);
2689      continue;
2690    }
2691
2692    // We aren't in the same group.  Are we descending into a new macro
2693    // or are part of an old one?
2694    std::shared_ptr<PathDiagnosticMacroPiece> MacroGroup;
2695
2696    SourceLocation ParentInstantiationLoc = InstantiationLoc.isMacroID() ?
2697                                          SM.getExpansionLoc(Loc) :
2698                                          SourceLocation();
2699
2700    // Walk the entire macro stack.
2701    while (!MacroStack.empty()) {
2702      if (InstantiationLoc == MacroStack.back().second) {
2703        MacroGroup = MacroStack.back().first;
2704        break;
2705      }
2706
2707      if (ParentInstantiationLoc == MacroStack.back().second) {
2708        MacroGroup = MacroStack.back().first;
2709        break;
2710      }
2711
2712      MacroStack.pop_back();
2713    }
2714
2715    if (!MacroGroup || ParentInstantiationLoc == MacroStack.back().second) {
2716      // Create a new macro group and add it to the stack.
2717      auto NewGroup = std::make_shared<PathDiagnosticMacroPiece>(
2718          PathDiagnosticLocation::createSingleLocation(piece->getLocation()));
2719
2720      if (MacroGroup)
2721        MacroGroup->subPieces.push_back(NewGroup);
2722      else {
2723        assert(InstantiationLoc.isFileID());
2724        Pieces.push_back(NewGroup);
2725      }
2726
2727      MacroGroup = NewGroup;
2728      MacroStack.push_back(std::make_pair(MacroGroup, InstantiationLoc));
2729    }
2730
2731    // Finally, add the PathDiagnosticPiece to the group.
2732    MacroGroup->subPieces.push_back(piece);
2733  }
2734
2735  // Now take the pieces and construct a new PathDiagnostic.
2736  path.clear();
2737
2738  path.insert(path.end(), Pieces.begin(), Pieces.end());
2739}
2740
2741/// Generate notes from all visitors.
2742/// Notes associated with {@code ErrorNode} are generated using
2743/// {@code getEndPath}, and the rest are generated with {@code VisitNode}.
2744static std::unique_ptr<VisitorsDiagnosticsTy>
2745generateVisitorsDiagnostics(PathSensitiveBugReport *R,
2746                            const ExplodedNode *ErrorNode,
2747                            BugReporterContext &BRC) {
2748  std::unique_ptr<VisitorsDiagnosticsTy> Notes =
2749      std::make_unique<VisitorsDiagnosticsTy>();
2750  PathSensitiveBugReport::VisitorList visitors;
2751
2752  // Run visitors on all nodes starting from the node *before* the last one.
2753  // The last node is reserved for notes generated with {@code getEndPath}.
2754  const ExplodedNode *NextNode = ErrorNode->getFirstPred();
2755  while (NextNode) {
2756
2757    // At each iteration, move all visitors from report to visitor list. This is
2758    // important, because the Profile() functions of the visitors make sure that
2759    // a visitor isn't added multiple times for the same node, but it's fine
2760    // to add the a visitor with Profile() for different nodes (e.g. tracking
2761    // a region at different points of the symbolic execution).
2762    for (std::unique_ptr<BugReporterVisitor> &Visitor : R->visitors())
2763      visitors.push_back(std::move(Visitor));
2764
2765    R->clearVisitors();
2766
2767    const ExplodedNode *Pred = NextNode->getFirstPred();
2768    if (!Pred) {
2769      PathDiagnosticPieceRef LastPiece;
2770      for (auto &V : visitors) {
2771        V->finalizeVisitor(BRC, ErrorNode, *R);
2772
2773        if (auto Piece = V->getEndPath(BRC, ErrorNode, *R)) {
2774          assert(!LastPiece &&
2775                 "There can only be one final piece in a diagnostic.");
2776          assert(Piece->getKind() == PathDiagnosticPiece::Kind::Event &&
2777                 "The final piece must contain a message!");
2778          LastPiece = std::move(Piece);
2779          (*Notes)[ErrorNode].push_back(LastPiece);
2780        }
2781      }
2782      break;
2783    }
2784
2785    for (auto &V : visitors) {
2786      auto P = V->VisitNode(NextNode, BRC, *R);
2787      if (P)
2788        (*Notes)[NextNode].push_back(std::move(P));
2789    }
2790
2791    if (!R->isValid())
2792      break;
2793
2794    NextNode = Pred;
2795  }
2796
2797  return Notes;
2798}
2799
2800Optional<PathDiagnosticBuilder> PathDiagnosticBuilder::findValidReport(
2801    ArrayRef<PathSensitiveBugReport *> &bugReports,
2802    PathSensitiveBugReporter &Reporter) {
2803
2804  BugPathGetter BugGraph(&Reporter.getGraph(), bugReports);
2805
2806  while (BugPathInfo *BugPath = BugGraph.getNextBugPath()) {
2807    // Find the BugReport with the original location.
2808    PathSensitiveBugReport *R = BugPath->Report;
2809    assert(R && "No original report found for sliced graph.");
2810    assert(R->isValid() && "Report selected by trimmed graph marked invalid.");
2811    const ExplodedNode *ErrorNode = BugPath->ErrorNode;
2812
2813    // Register refutation visitors first, if they mark the bug invalid no
2814    // further analysis is required
2815    R->addVisitor(std::make_unique<LikelyFalsePositiveSuppressionBRVisitor>());
2816
2817    // Register additional node visitors.
2818    R->addVisitor(std::make_unique<NilReceiverBRVisitor>());
2819    R->addVisitor(std::make_unique<ConditionBRVisitor>());
2820    R->addVisitor(std::make_unique<TagVisitor>());
2821
2822    BugReporterContext BRC(Reporter);
2823
2824    // Run all visitors on a given graph, once.
2825    std::unique_ptr<VisitorsDiagnosticsTy> visitorNotes =
2826        generateVisitorsDiagnostics(R, ErrorNode, BRC);
2827
2828    if (R->isValid()) {
2829      if (Reporter.getAnalyzerOptions().ShouldCrosscheckWithZ3) {
2830        // If crosscheck is enabled, remove all visitors, add the refutation
2831        // visitor and check again
2832        R->clearVisitors();
2833        R->addVisitor(std::make_unique<FalsePositiveRefutationBRVisitor>());
2834
2835        // We don't overwrite the notes inserted by other visitors because the
2836        // refutation manager does not add any new note to the path
2837        generateVisitorsDiagnostics(R, BugPath->ErrorNode, BRC);
2838      }
2839
2840      // Check if the bug is still valid
2841      if (R->isValid())
2842        return PathDiagnosticBuilder(
2843            std::move(BRC), std::move(BugPath->BugPath), BugPath->Report,
2844            BugPath->ErrorNode, std::move(visitorNotes));
2845    }
2846  }
2847
2848  return {};
2849}
2850
2851std::unique_ptr<DiagnosticForConsumerMapTy>
2852PathSensitiveBugReporter::generatePathDiagnostics(
2853    ArrayRef<PathDiagnosticConsumer *> consumers,
2854    ArrayRef<PathSensitiveBugReport *> &bugReports) {
2855  assert(!bugReports.empty());
2856
2857  auto Out = std::make_unique<DiagnosticForConsumerMapTy>();
2858
2859  Optional<PathDiagnosticBuilder> PDB =
2860      PathDiagnosticBuilder::findValidReport(bugReports, *this);
2861
2862  if (PDB) {
2863    for (PathDiagnosticConsumer *PC : consumers) {
2864      if (std::unique_ptr<PathDiagnostic> PD = PDB->generate(PC)) {
2865        (*Out)[PC] = std::move(PD);
2866      }
2867    }
2868  }
2869
2870  return Out;
2871}
2872
2873void BugReporter::emitReport(std::unique_ptr<BugReport> R) {
2874  bool ValidSourceLoc = R->getLocation().isValid();
2875  assert(ValidSourceLoc);
2876  // If we mess up in a release build, we'd still prefer to just drop the bug
2877  // instead of trying to go on.
2878  if (!ValidSourceLoc)
2879    return;
2880
2881  // Compute the bug report's hash to determine its equivalence class.
2882  llvm::FoldingSetNodeID ID;
2883  R->Profile(ID);
2884
2885  // Lookup the equivance class.  If there isn't one, create it.
2886  void *InsertPos;
2887  BugReportEquivClass* EQ = EQClasses.FindNodeOrInsertPos(ID, InsertPos);
2888
2889  if (!EQ) {
2890    EQ = new BugReportEquivClass(std::move(R));
2891    EQClasses.InsertNode(EQ, InsertPos);
2892    EQClassesVector.push_back(EQ);
2893  } else
2894    EQ->AddReport(std::move(R));
2895}
2896
2897void PathSensitiveBugReporter::emitReport(std::unique_ptr<BugReport> R) {
2898  if (auto PR = dyn_cast<PathSensitiveBugReport>(R.get()))
2899    if (const ExplodedNode *E = PR->getErrorNode()) {
2900      // An error node must either be a sink or have a tag, otherwise
2901      // it could get reclaimed before the path diagnostic is created.
2902      assert((E->isSink() || E->getLocation().getTag()) &&
2903             "Error node must either be a sink or have a tag");
2904
2905      const AnalysisDeclContext *DeclCtx =
2906          E->getLocationContext()->getAnalysisDeclContext();
2907      // The source of autosynthesized body can be handcrafted AST or a model
2908      // file. The locations from handcrafted ASTs have no valid source
2909      // locations and have to be discarded. Locations from model files should
2910      // be preserved for processing and reporting.
2911      if (DeclCtx->isBodyAutosynthesized() &&
2912          !DeclCtx->isBodyAutosynthesizedFromModelFile())
2913        return;
2914    }
2915
2916  BugReporter::emitReport(std::move(R));
2917}
2918
2919//===----------------------------------------------------------------------===//
2920// Emitting reports in equivalence classes.
2921//===----------------------------------------------------------------------===//
2922
2923namespace {
2924
2925struct FRIEC_WLItem {
2926  const ExplodedNode *N;
2927  ExplodedNode::const_succ_iterator I, E;
2928
2929  FRIEC_WLItem(const ExplodedNode *n)
2930      : N(n), I(N->succ_begin()), E(N->succ_end()) {}
2931};
2932
2933} // namespace
2934
2935BugReport *PathSensitiveBugReporter::findReportInEquivalenceClass(
2936    BugReportEquivClass &EQ, SmallVectorImpl<BugReport *> &bugReports) {
2937  // If we don't need to suppress any of the nodes because they are
2938  // post-dominated by a sink, simply add all the nodes in the equivalence class
2939  // to 'Nodes'.  Any of the reports will serve as a "representative" report.
2940  assert(EQ.getReports().size() > 0);
2941  const BugType& BT = EQ.getReports()[0]->getBugType();
2942  if (!BT.isSuppressOnSink()) {
2943    BugReport *R = EQ.getReports()[0].get();
2944    for (auto &J : EQ.getReports()) {
2945      if (auto *PR = dyn_cast<PathSensitiveBugReport>(J.get())) {
2946        R = PR;
2947        bugReports.push_back(PR);
2948      }
2949    }
2950    return R;
2951  }
2952
2953  // For bug reports that should be suppressed when all paths are post-dominated
2954  // by a sink node, iterate through the reports in the equivalence class
2955  // until we find one that isn't post-dominated (if one exists).  We use a
2956  // DFS traversal of the ExplodedGraph to find a non-sink node.  We could write
2957  // this as a recursive function, but we don't want to risk blowing out the
2958  // stack for very long paths.
2959  BugReport *exampleReport = nullptr;
2960
2961  for (const auto &I: EQ.getReports()) {
2962    auto *R = dyn_cast<PathSensitiveBugReport>(I.get());
2963    if (!R)
2964      continue;
2965
2966    const ExplodedNode *errorNode = R->getErrorNode();
2967    if (errorNode->isSink()) {
2968      llvm_unreachable(
2969           "BugType::isSuppressSink() should not be 'true' for sink end nodes");
2970    }
2971    // No successors?  By definition this nodes isn't post-dominated by a sink.
2972    if (errorNode->succ_empty()) {
2973      bugReports.push_back(R);
2974      if (!exampleReport)
2975        exampleReport = R;
2976      continue;
2977    }
2978
2979    // See if we are in a no-return CFG block. If so, treat this similarly
2980    // to being post-dominated by a sink. This works better when the analysis
2981    // is incomplete and we have never reached the no-return function call(s)
2982    // that we'd inevitably bump into on this path.
2983    if (const CFGBlock *ErrorB = errorNode->getCFGBlock())
2984      if (ErrorB->isInevitablySinking())
2985        continue;
2986
2987    // At this point we know that 'N' is not a sink and it has at least one
2988    // successor.  Use a DFS worklist to find a non-sink end-of-path node.
2989    using WLItem = FRIEC_WLItem;
2990    using DFSWorkList = SmallVector<WLItem, 10>;
2991
2992    llvm::DenseMap<const ExplodedNode *, unsigned> Visited;
2993
2994    DFSWorkList WL;
2995    WL.push_back(errorNode);
2996    Visited[errorNode] = 1;
2997
2998    while (!WL.empty()) {
2999      WLItem &WI = WL.back();
3000      assert(!WI.N->succ_empty());
3001
3002      for (; WI.I != WI.E; ++WI.I) {
3003        const ExplodedNode *Succ = *WI.I;
3004        // End-of-path node?
3005        if (Succ->succ_empty()) {
3006          // If we found an end-of-path node that is not a sink.
3007          if (!Succ->isSink()) {
3008            bugReports.push_back(R);
3009            if (!exampleReport)
3010              exampleReport = R;
3011            WL.clear();
3012            break;
3013          }
3014          // Found a sink?  Continue on to the next successor.
3015          continue;
3016        }
3017        // Mark the successor as visited.  If it hasn't been explored,
3018        // enqueue it to the DFS worklist.
3019        unsigned &mark = Visited[Succ];
3020        if (!mark) {
3021          mark = 1;
3022          WL.push_back(Succ);
3023          break;
3024        }
3025      }
3026
3027      // The worklist may have been cleared at this point.  First
3028      // check if it is empty before checking the last item.
3029      if (!WL.empty() && &WL.back() == &WI)
3030        WL.pop_back();
3031    }
3032  }
3033
3034  // ExampleReport will be NULL if all the nodes in the equivalence class
3035  // were post-dominated by sinks.
3036  return exampleReport;
3037}
3038
3039void BugReporter::FlushReport(BugReportEquivClass& EQ) {
3040  SmallVector<BugReport*, 10> bugReports;
3041  BugReport *report = findReportInEquivalenceClass(EQ, bugReports);
3042  if (!report)
3043    return;
3044
3045  ArrayRef<PathDiagnosticConsumer*> Consumers = getPathDiagnosticConsumers();
3046  std::unique_ptr<DiagnosticForConsumerMapTy> Diagnostics =
3047      generateDiagnosticForConsumerMap(report, Consumers, bugReports);
3048
3049  for (auto &P : *Diagnostics) {
3050    PathDiagnosticConsumer *Consumer = P.first;
3051    std::unique_ptr<PathDiagnostic> &PD = P.second;
3052
3053    // If the path is empty, generate a single step path with the location
3054    // of the issue.
3055    if (PD->path.empty()) {
3056      PathDiagnosticLocation L = report->getLocation();
3057      auto piece = std::make_unique<PathDiagnosticEventPiece>(
3058        L, report->getDescription());
3059      for (SourceRange Range : report->getRanges())
3060        piece->addRange(Range);
3061      PD->setEndOfPath(std::move(piece));
3062    }
3063
3064    PathPieces &Pieces = PD->getMutablePieces();
3065    if (getAnalyzerOptions().ShouldDisplayNotesAsEvents) {
3066      // For path diagnostic consumers that don't support extra notes,
3067      // we may optionally convert those to path notes.
3068      for (auto I = report->getNotes().rbegin(),
3069           E = report->getNotes().rend(); I != E; ++I) {
3070        PathDiagnosticNotePiece *Piece = I->get();
3071        auto ConvertedPiece = std::make_shared<PathDiagnosticEventPiece>(
3072          Piece->getLocation(), Piece->getString());
3073        for (const auto &R: Piece->getRanges())
3074          ConvertedPiece->addRange(R);
3075
3076        Pieces.push_front(std::move(ConvertedPiece));
3077      }
3078    } else {
3079      for (auto I = report->getNotes().rbegin(),
3080           E = report->getNotes().rend(); I != E; ++I)
3081        Pieces.push_front(*I);
3082    }
3083
3084    for (const auto &I : report->getFixits())
3085      Pieces.back()->addFixit(I);
3086
3087    updateExecutedLinesWithDiagnosticPieces(*PD);
3088    Consumer->HandlePathDiagnostic(std::move(PD));
3089  }
3090}
3091
3092/// Insert all lines participating in the function signature \p Signature
3093/// into \p ExecutedLines.
3094static void populateExecutedLinesWithFunctionSignature(
3095    const Decl *Signature, const SourceManager &SM,
3096    FilesToLineNumsMap &ExecutedLines) {
3097  SourceRange SignatureSourceRange;
3098  const Stmt* Body = Signature->getBody();
3099  if (const auto FD = dyn_cast<FunctionDecl>(Signature)) {
3100    SignatureSourceRange = FD->getSourceRange();
3101  } else if (const auto OD = dyn_cast<ObjCMethodDecl>(Signature)) {
3102    SignatureSourceRange = OD->getSourceRange();
3103  } else {
3104    return;
3105  }
3106  SourceLocation Start = SignatureSourceRange.getBegin();
3107  SourceLocation End = Body ? Body->getSourceRange().getBegin()
3108    : SignatureSourceRange.getEnd();
3109  if (!Start.isValid() || !End.isValid())
3110    return;
3111  unsigned StartLine = SM.getExpansionLineNumber(Start);
3112  unsigned EndLine = SM.getExpansionLineNumber(End);
3113
3114  FileID FID = SM.getFileID(SM.getExpansionLoc(Start));
3115  for (unsigned Line = StartLine; Line <= EndLine; Line++)
3116    ExecutedLines[FID].insert(Line);
3117}
3118
3119static void populateExecutedLinesWithStmt(
3120    const Stmt *S, const SourceManager &SM,
3121    FilesToLineNumsMap &ExecutedLines) {
3122  SourceLocation Loc = S->getSourceRange().getBegin();
3123  if (!Loc.isValid())
3124    return;
3125  SourceLocation ExpansionLoc = SM.getExpansionLoc(Loc);
3126  FileID FID = SM.getFileID(ExpansionLoc);
3127  unsigned LineNo = SM.getExpansionLineNumber(ExpansionLoc);
3128  ExecutedLines[FID].insert(LineNo);
3129}
3130
3131/// \return all executed lines including function signatures on the path
3132/// starting from \p N.
3133static std::unique_ptr<FilesToLineNumsMap>
3134findExecutedLines(const SourceManager &SM, const ExplodedNode *N) {
3135  auto ExecutedLines = std::make_unique<FilesToLineNumsMap>();
3136
3137  while (N) {
3138    if (N->getFirstPred() == nullptr) {
3139      // First node: show signature of the entrance point.
3140      const Decl *D = N->getLocationContext()->getDecl();
3141      populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines);
3142    } else if (auto CE = N->getLocationAs<CallEnter>()) {
3143      // Inlined function: show signature.
3144      const Decl* D = CE->getCalleeContext()->getDecl();
3145      populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines);
3146    } else if (const Stmt *S = N->getStmtForDiagnostics()) {
3147      populateExecutedLinesWithStmt(S, SM, *ExecutedLines);
3148
3149      // Show extra context for some parent kinds.
3150      const Stmt *P = N->getParentMap().getParent(S);
3151
3152      // The path exploration can die before the node with the associated
3153      // return statement is generated, but we do want to show the whole
3154      // return.
3155      if (const auto *RS = dyn_cast_or_null<ReturnStmt>(P)) {
3156        populateExecutedLinesWithStmt(RS, SM, *ExecutedLines);
3157        P = N->getParentMap().getParent(RS);
3158      }
3159
3160      if (P && (isa<SwitchCase>(P) || isa<LabelStmt>(P)))
3161        populateExecutedLinesWithStmt(P, SM, *ExecutedLines);
3162    }
3163
3164    N = N->getFirstPred();
3165  }
3166  return ExecutedLines;
3167}
3168
3169std::unique_ptr<DiagnosticForConsumerMapTy>
3170BugReporter::generateDiagnosticForConsumerMap(
3171    BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers,
3172    ArrayRef<BugReport *> bugReports) {
3173  auto *basicReport = cast<BasicBugReport>(exampleReport);
3174  auto Out = std::make_unique<DiagnosticForConsumerMapTy>();
3175  for (auto *Consumer : consumers)
3176    (*Out)[Consumer] = generateDiagnosticForBasicReport(basicReport);
3177  return Out;
3178}
3179
3180static PathDiagnosticCallPiece *
3181getFirstStackedCallToHeaderFile(PathDiagnosticCallPiece *CP,
3182                                const SourceManager &SMgr) {
3183  SourceLocation CallLoc = CP->callEnter.asLocation();
3184
3185  // If the call is within a macro, don't do anything (for now).
3186  if (CallLoc.isMacroID())
3187    return nullptr;
3188
3189  assert(AnalysisManager::isInCodeFile(CallLoc, SMgr) &&
3190         "The call piece should not be in a header file.");
3191
3192  // Check if CP represents a path through a function outside of the main file.
3193  if (!AnalysisManager::isInCodeFile(CP->callEnterWithin.asLocation(), SMgr))
3194    return CP;
3195
3196  const PathPieces &Path = CP->path;
3197  if (Path.empty())
3198    return nullptr;
3199
3200  // Check if the last piece in the callee path is a call to a function outside
3201  // of the main file.
3202  if (auto *CPInner = dyn_cast<PathDiagnosticCallPiece>(Path.back().get()))
3203    return getFirstStackedCallToHeaderFile(CPInner, SMgr);
3204
3205  // Otherwise, the last piece is in the main file.
3206  return nullptr;
3207}
3208
3209static void resetDiagnosticLocationToMainFile(PathDiagnostic &PD) {
3210  if (PD.path.empty())
3211    return;
3212
3213  PathDiagnosticPiece *LastP = PD.path.back().get();
3214  assert(LastP);
3215  const SourceManager &SMgr = LastP->getLocation().getManager();
3216
3217  // We only need to check if the report ends inside headers, if the last piece
3218  // is a call piece.
3219  if (auto *CP = dyn_cast<PathDiagnosticCallPiece>(LastP)) {
3220    CP = getFirstStackedCallToHeaderFile(CP, SMgr);
3221    if (CP) {
3222      // Mark the piece.
3223       CP->setAsLastInMainSourceFile();
3224
3225      // Update the path diagnostic message.
3226      const auto *ND = dyn_cast<NamedDecl>(CP->getCallee());
3227      if (ND) {
3228        SmallString<200> buf;
3229        llvm::raw_svector_ostream os(buf);
3230        os << " (within a call to '" << ND->getDeclName() << "')";
3231        PD.appendToDesc(os.str());
3232      }
3233
3234      // Reset the report containing declaration and location.
3235      PD.setDeclWithIssue(CP->getCaller());
3236      PD.setLocation(CP->getLocation());
3237
3238      return;
3239    }
3240  }
3241}
3242
3243
3244
3245std::unique_ptr<DiagnosticForConsumerMapTy>
3246PathSensitiveBugReporter::generateDiagnosticForConsumerMap(
3247    BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers,
3248    ArrayRef<BugReport *> bugReports) {
3249  std::vector<BasicBugReport *> BasicBugReports;
3250  std::vector<PathSensitiveBugReport *> PathSensitiveBugReports;
3251  if (isa<BasicBugReport>(exampleReport))
3252    return BugReporter::generateDiagnosticForConsumerMap(exampleReport,
3253                                                         consumers, bugReports);
3254
3255  // Generate the full path sensitive diagnostic, using the generation scheme
3256  // specified by the PathDiagnosticConsumer. Note that we have to generate
3257  // path diagnostics even for consumers which do not support paths, because
3258  // the BugReporterVisitors may mark this bug as a false positive.
3259  assert(!bugReports.empty());
3260  MaxBugClassSize.updateMax(bugReports.size());
3261
3262  // Avoid copying the whole array because there may be a lot of reports.
3263  ArrayRef<PathSensitiveBugReport *> convertedArrayOfReports(
3264      reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.begin()),
3265      reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.end()));
3266  std::unique_ptr<DiagnosticForConsumerMapTy> Out = generatePathDiagnostics(
3267      consumers, convertedArrayOfReports);
3268
3269  if (Out->empty())
3270    return Out;
3271
3272  MaxValidBugClassSize.updateMax(bugReports.size());
3273
3274  // Examine the report and see if the last piece is in a header. Reset the
3275  // report location to the last piece in the main source file.
3276  const AnalyzerOptions &Opts = getAnalyzerOptions();
3277  for (auto const &P : *Out)
3278    if (Opts.ShouldReportIssuesInMainSourceFile && !Opts.AnalyzeAll)
3279      resetDiagnosticLocationToMainFile(*P.second);
3280
3281  return Out;
3282}
3283
3284void BugReporter::EmitBasicReport(const Decl *DeclWithIssue,
3285                                  const CheckerBase *Checker, StringRef Name,
3286                                  StringRef Category, StringRef Str,
3287                                  PathDiagnosticLocation Loc,
3288                                  ArrayRef<SourceRange> Ranges,
3289                                  ArrayRef<FixItHint> Fixits) {
3290  EmitBasicReport(DeclWithIssue, Checker->getCheckerName(), Name, Category, Str,
3291                  Loc, Ranges, Fixits);
3292}
3293
3294void BugReporter::EmitBasicReport(const Decl *DeclWithIssue,
3295                                  CheckerNameRef CheckName,
3296                                  StringRef name, StringRef category,
3297                                  StringRef str, PathDiagnosticLocation Loc,
3298                                  ArrayRef<SourceRange> Ranges,
3299                                  ArrayRef<FixItHint> Fixits) {
3300  // 'BT' is owned by BugReporter.
3301  BugType *BT = getBugTypeForName(CheckName, name, category);
3302  auto R = std::make_unique<BasicBugReport>(*BT, str, Loc);
3303  R->setDeclWithIssue(DeclWithIssue);
3304  for (const auto &SR : Ranges)
3305    R->addRange(SR);
3306  for (const auto &FH : Fixits)
3307    R->addFixItHint(FH);
3308  emitReport(std::move(R));
3309}
3310
3311BugType *BugReporter::getBugTypeForName(CheckerNameRef CheckName,
3312                                        StringRef name, StringRef category) {
3313  SmallString<136> fullDesc;
3314  llvm::raw_svector_ostream(fullDesc) << CheckName.getName() << ":" << name
3315                                      << ":" << category;
3316  std::unique_ptr<BugType> &BT = StrBugTypes[fullDesc];
3317  if (!BT)
3318    BT = std::make_unique<BugType>(CheckName, name, category);
3319  return BT.get();
3320}
3321