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 shouldAddControlNotes() const {
192 return Consumer->shouldAddControlNotes();
193 }
194 bool shouldGenerateDiagnostics() const {
195 return Consumer->shouldGenerateDiagnostics();
196 }
197 bool supportsLogicalOpControlFlow() const {
198 return Consumer->supportsLogicalOpControlFlow();
199 }
200};
201
202/// Contains every contextual information needed for constructing a
203/// PathDiagnostic object for a given bug report. This class and its fields are
204/// immutable, and passes a BugReportConstruct object around during the
205/// construction.
206class PathDiagnosticBuilder : public BugReporterContext {
207 /// A linear path from the error node to the root.
208 std::unique_ptr<const ExplodedGraph> BugPath;
209 /// The bug report we're describing. Visitors create their diagnostics with
210 /// them being the last entities being able to modify it (for example,
211 /// changing interestingness here would cause inconsistencies as to how this
212 /// file and visitors construct diagnostics), hence its const.
213 const PathSensitiveBugReport *R;
214 /// The leaf of the bug path. This isn't the same as the bug reports error
215 /// node, which refers to the *original* graph, not the bug path.
216 const ExplodedNode *const ErrorNode;
217 /// The diagnostic pieces visitors emitted, which is expected to be collected
218 /// by the time this builder is constructed.
219 std::unique_ptr<const VisitorsDiagnosticsTy> VisitorsDiagnostics;
220
221public:
222 /// Find a non-invalidated report for a given equivalence class, and returns
223 /// a PathDiagnosticBuilder able to construct bug reports for different
224 /// consumers. Returns None if no valid report is found.
225 static Optional<PathDiagnosticBuilder>
226 findValidReport(ArrayRef<PathSensitiveBugReport *> &bugReports,
227 PathSensitiveBugReporter &Reporter);
228
229 PathDiagnosticBuilder(
230 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath,
231 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode,
232 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics);
233
234 /// This function is responsible for generating diagnostic pieces that are
235 /// *not* provided by bug report visitors.
236 /// These diagnostics may differ depending on the consumer's settings,
237 /// and are therefore constructed separately for each consumer.
238 ///
239 /// There are two path diagnostics generation modes: with adding edges (used
240 /// for plists) and without (used for HTML and text). When edges are added,
241 /// the path is modified to insert artificially generated edges.
242 /// Otherwise, more detailed diagnostics is emitted for block edges,
243 /// explaining the transitions in words.
244 std::unique_ptr<PathDiagnostic>
245 generate(const PathDiagnosticConsumer *PDC) const;
246
247private:
248 void updateStackPiecesWithMessage(PathDiagnosticPieceRef P,
249 const CallWithEntryStack &CallStack) const;
250 void generatePathDiagnosticsForNode(PathDiagnosticConstruct &C,
251 PathDiagnosticLocation &PrevLoc) const;
252
253 void generateMinimalDiagForBlockEdge(PathDiagnosticConstruct &C,
254 BlockEdge BE) const;
255
256 PathDiagnosticPieceRef
257 generateDiagForGotoOP(const PathDiagnosticConstruct &C, const Stmt *S,
258 PathDiagnosticLocation &Start) const;
259
260 PathDiagnosticPieceRef
261 generateDiagForSwitchOP(const PathDiagnosticConstruct &C, const CFGBlock *Dst,
262 PathDiagnosticLocation &Start) const;
263
264 PathDiagnosticPieceRef
265 generateDiagForBinaryOP(const PathDiagnosticConstruct &C, const Stmt *T,
266 const CFGBlock *Src, const CFGBlock *DstC) const;
267
268 PathDiagnosticLocation
269 ExecutionContinues(const PathDiagnosticConstruct &C) const;
270
271 PathDiagnosticLocation
272 ExecutionContinues(llvm::raw_string_ostream &os,
273 const PathDiagnosticConstruct &C) const;
274
275 const PathSensitiveBugReport *getBugReport() const { return R; }
276};
277
278} // namespace
279
280//===----------------------------------------------------------------------===//
281// Base implementation of stack hint generators.
282//===----------------------------------------------------------------------===//
283
284StackHintGenerator::~StackHintGenerator() = default;
285
286std::string StackHintGeneratorForSymbol::getMessage(const ExplodedNode *N){
287 if (!N)
288 return getMessageForSymbolNotFound();
289
290 ProgramPoint P = N->getLocation();
291 CallExitEnd CExit = P.castAs<CallExitEnd>();
292
293 // FIXME: Use CallEvent to abstract this over all calls.
294 const Stmt *CallSite = CExit.getCalleeContext()->getCallSite();
295 const auto *CE = dyn_cast_or_null<CallExpr>(CallSite);
296 if (!CE)
297 return {};
298
299 // Check if one of the parameters are set to the interesting symbol.
300 unsigned ArgIndex = 0;
301 for (CallExpr::const_arg_iterator I = CE->arg_begin(),
302 E = CE->arg_end(); I != E; ++I, ++ArgIndex){
303 SVal SV = N->getSVal(*I);
304
305 // Check if the variable corresponding to the symbol is passed by value.
306 SymbolRef AS = SV.getAsLocSymbol();
307 if (AS == Sym) {
308 return getMessageForArg(*I, ArgIndex);
309 }
310
311 // Check if the parameter is a pointer to the symbol.
312 if (Optional<loc::MemRegionVal> Reg = SV.getAs<loc::MemRegionVal>()) {
313 // Do not attempt to dereference void*.
314 if ((*I)->getType()->isVoidPointerType())
315 continue;
316 SVal PSV = N->getState()->getSVal(Reg->getRegion());
317 SymbolRef AS = PSV.getAsLocSymbol();
318 if (AS == Sym) {
319 return getMessageForArg(*I, ArgIndex);
320 }
321 }
322 }
323
324 // Check if we are returning the interesting symbol.
325 SVal SV = N->getSVal(CE);
326 SymbolRef RetSym = SV.getAsLocSymbol();
327 if (RetSym == Sym) {
328 return getMessageForReturn(CE);
329 }
330
331 return getMessageForSymbolNotFound();
332}
333
334std::string StackHintGeneratorForSymbol::getMessageForArg(const Expr *ArgE,
335 unsigned ArgIndex) {
336 // Printed parameters start at 1, not 0.
337 ++ArgIndex;
338
339 return (llvm::Twine(Msg) + " via " + std::to_string(ArgIndex) +
340 llvm::getOrdinalSuffix(ArgIndex) + " parameter").str();
341}
342
343//===----------------------------------------------------------------------===//
344// Diagnostic cleanup.
345//===----------------------------------------------------------------------===//
346
347static PathDiagnosticEventPiece *
348eventsDescribeSameCondition(PathDiagnosticEventPiece *X,
349 PathDiagnosticEventPiece *Y) {
350 // Prefer diagnostics that come from ConditionBRVisitor over
351 // those that came from TrackConstraintBRVisitor,
352 // unless the one from ConditionBRVisitor is
353 // its generic fallback diagnostic.
354 const void *tagPreferred = ConditionBRVisitor::getTag();
355 const void *tagLesser = TrackConstraintBRVisitor::getTag();
356
357 if (X->getLocation() != Y->getLocation())
358 return nullptr;
359
360 if (X->getTag() == tagPreferred && Y->getTag() == tagLesser)
361 return ConditionBRVisitor::isPieceMessageGeneric(X) ? Y : X;
362
363 if (Y->getTag() == tagPreferred && X->getTag() == tagLesser)
364 return ConditionBRVisitor::isPieceMessageGeneric(Y) ? X : Y;
365
366 return nullptr;
367}
368
369/// An optimization pass over PathPieces that removes redundant diagnostics
370/// generated by both ConditionBRVisitor and TrackConstraintBRVisitor. Both
371/// BugReporterVisitors use different methods to generate diagnostics, with
372/// one capable of emitting diagnostics in some cases but not in others. This
373/// can lead to redundant diagnostic pieces at the same point in a path.
374static void removeRedundantMsgs(PathPieces &path) {
375 unsigned N = path.size();
376 if (N < 2)
377 return;
378 // NOTE: this loop intentionally is not using an iterator. Instead, we
379 // are streaming the path and modifying it in place. This is done by
380 // grabbing the front, processing it, and if we decide to keep it append
381 // it to the end of the path. The entire path is processed in this way.
382 for (unsigned i = 0; i < N; ++i) {
383 auto piece = std::move(path.front());
384 path.pop_front();
385
386 switch (piece->getKind()) {
387 case PathDiagnosticPiece::Call:
388 removeRedundantMsgs(cast<PathDiagnosticCallPiece>(*piece).path);
389 break;
390 case PathDiagnosticPiece::Macro:
391 removeRedundantMsgs(cast<PathDiagnosticMacroPiece>(*piece).subPieces);
392 break;
393 case PathDiagnosticPiece::Event: {
394 if (i == N-1)
395 break;
396
397 if (auto *nextEvent =
398 dyn_cast<PathDiagnosticEventPiece>(path.front().get())) {
399 auto *event = cast<PathDiagnosticEventPiece>(piece.get());
400 // Check to see if we should keep one of the two pieces. If we
401 // come up with a preference, record which piece to keep, and consume
402 // another piece from the path.
403 if (auto *pieceToKeep =
404 eventsDescribeSameCondition(event, nextEvent)) {
405 piece = std::move(pieceToKeep == event ? piece : path.front());
406 path.pop_front();
407 ++i;
408 }
409 }
410 break;
411 }
412 case PathDiagnosticPiece::ControlFlow:
413 case PathDiagnosticPiece::Note:
414 case PathDiagnosticPiece::PopUp:
415 break;
416 }
417 path.push_back(std::move(piece));
418 }
419}
420
421/// Recursively scan through a path and prune out calls and macros pieces
422/// that aren't needed. Return true if afterwards the path contains
423/// "interesting stuff" which means it shouldn't be pruned from the parent path.
424static bool removeUnneededCalls(const PathDiagnosticConstruct &C,
425 PathPieces &pieces,
426 const PathSensitiveBugReport *R,
427 bool IsInteresting = false) {
428 bool containsSomethingInteresting = IsInteresting;
429 const unsigned N = pieces.size();
430
431 for (unsigned i = 0 ; i < N ; ++i) {
432 // Remove the front piece from the path. If it is still something we
433 // want to keep once we are done, we will push it back on the end.
434 auto piece = std::move(pieces.front());
435 pieces.pop_front();
436
437 switch (piece->getKind()) {
438 case PathDiagnosticPiece::Call: {
439 auto &call = cast<PathDiagnosticCallPiece>(*piece);
440 // Check if the location context is interesting.
441 if (!removeUnneededCalls(
442 C, call.path, R,
443 R->isInteresting(C.getLocationContextFor(&call.path))))
444 continue;
445
446 containsSomethingInteresting = true;
447 break;
448 }
449 case PathDiagnosticPiece::Macro: {
450 auto &macro = cast<PathDiagnosticMacroPiece>(*piece);
451 if (!removeUnneededCalls(C, macro.subPieces, R, IsInteresting))
452 continue;
453 containsSomethingInteresting = true;
454 break;
455 }
456 case PathDiagnosticPiece::Event: {
457 auto &event = cast<PathDiagnosticEventPiece>(*piece);
458
459 // We never throw away an event, but we do throw it away wholesale
460 // as part of a path if we throw the entire path away.
461 containsSomethingInteresting |= !event.isPrunable();
462 break;
463 }
464 case PathDiagnosticPiece::ControlFlow:
465 case PathDiagnosticPiece::Note:
466 case PathDiagnosticPiece::PopUp:
467 break;
468 }
469
470 pieces.push_back(std::move(piece));
471 }
472
473 return containsSomethingInteresting;
474}
475
476/// Same logic as above to remove extra pieces.
477static void removePopUpNotes(PathPieces &Path) {
478 for (unsigned int i = 0; i < Path.size(); ++i) {
479 auto Piece = std::move(Path.front());
480 Path.pop_front();
481 if (!isa<PathDiagnosticPopUpPiece>(*Piece))
482 Path.push_back(std::move(Piece));
483 }
484}
485
486/// Returns true if the given decl has been implicitly given a body, either by
487/// the analyzer or by the compiler proper.
488static bool hasImplicitBody(const Decl *D) {
489 assert(D);
490 return D->isImplicit() || !D->hasBody();
491}
492
493/// Recursively scan through a path and make sure that all call pieces have
494/// valid locations.
495static void
496adjustCallLocations(PathPieces &Pieces,
497 PathDiagnosticLocation *LastCallLocation = nullptr) {
498 for (const auto &I : Pieces) {
499 auto *Call = dyn_cast<PathDiagnosticCallPiece>(I.get());
500
501 if (!Call)
502 continue;
503
504 if (LastCallLocation) {
505 bool CallerIsImplicit = hasImplicitBody(Call->getCaller());
506 if (CallerIsImplicit || !Call->callEnter.asLocation().isValid())
507 Call->callEnter = *LastCallLocation;
508 if (CallerIsImplicit || !Call->callReturn.asLocation().isValid())
509 Call->callReturn = *LastCallLocation;
510 }
511
512 // Recursively clean out the subclass. Keep this call around if
513 // it contains any informative diagnostics.
514 PathDiagnosticLocation *ThisCallLocation;
515 if (Call->callEnterWithin.asLocation().isValid() &&
516 !hasImplicitBody(Call->getCallee()))
517 ThisCallLocation = &Call->callEnterWithin;
518 else
519 ThisCallLocation = &Call->callEnter;
520
521 assert(ThisCallLocation && "Outermost call has an invalid location");
522 adjustCallLocations(Call->path, ThisCallLocation);
523 }
524}
525
526/// Remove edges in and out of C++ default initializer expressions. These are
527/// for fields that have in-class initializers, as opposed to being initialized
528/// explicitly in a constructor or braced list.
529static void removeEdgesToDefaultInitializers(PathPieces &Pieces) {
530 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) {
531 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get()))
532 removeEdgesToDefaultInitializers(C->path);
533
534 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get()))
535 removeEdgesToDefaultInitializers(M->subPieces);
536
537 if (auto *CF = dyn_cast<PathDiagnosticControlFlowPiece>(I->get())) {
538 const Stmt *Start = CF->getStartLocation().asStmt();
539 const Stmt *End = CF->getEndLocation().asStmt();
540 if (isa_and_nonnull<CXXDefaultInitExpr>(Start)) {
541 I = Pieces.erase(I);
542 continue;
543 } else if (isa_and_nonnull<CXXDefaultInitExpr>(End)) {
544 PathPieces::iterator Next = std::next(I);
545 if (Next != E) {
546 if (auto *NextCF =
547 dyn_cast<PathDiagnosticControlFlowPiece>(Next->get())) {
548 NextCF->setStartLocation(CF->getStartLocation());
549 }
550 }
551 I = Pieces.erase(I);
552 continue;
553 }
554 }
555
556 I++;
557 }
558}
559
560/// Remove all pieces with invalid locations as these cannot be serialized.
561/// We might have pieces with invalid locations as a result of inlining Body
562/// Farm generated functions.
563static void removePiecesWithInvalidLocations(PathPieces &Pieces) {
564 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) {
565 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get()))
566 removePiecesWithInvalidLocations(C->path);
567
568 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get()))
569 removePiecesWithInvalidLocations(M->subPieces);
570
571 if (!(*I)->getLocation().isValid() ||
572 !(*I)->getLocation().asLocation().isValid()) {
573 I = Pieces.erase(I);
574 continue;
575 }
576 I++;
577 }
578}
579
580PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues(
581 const PathDiagnosticConstruct &C) const {
582 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics())
583 return PathDiagnosticLocation(S, getSourceManager(),
584 C.getCurrLocationContext());
585
586 return PathDiagnosticLocation::createDeclEnd(C.getCurrLocationContext(),
587 getSourceManager());
588}
589
590PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues(
591 llvm::raw_string_ostream &os, const PathDiagnosticConstruct &C) const {
592 // Slow, but probably doesn't matter.
593 if (os.str().empty())
594 os << ' ';
595
596 const PathDiagnosticLocation &Loc = ExecutionContinues(C);
597
598 if (Loc.asStmt())
599 os << "Execution continues on line "
600 << getSourceManager().getExpansionLineNumber(Loc.asLocation())
601 << '.';
602 else {
603 os << "Execution jumps to the end of the ";
604 const Decl *D = C.getCurrLocationContext()->getDecl();
605 if (isa<ObjCMethodDecl>(D))
606 os << "method";
607 else if (isa<FunctionDecl>(D))
608 os << "function";
609 else {
610 assert(isa<BlockDecl>(D));
611 os << "anonymous block";
612 }
613 os << '.';
614 }
615
616 return Loc;
617}
618
619static const Stmt *getEnclosingParent(const Stmt *S, const ParentMap &PM) {
620 if (isa<Expr>(S) && PM.isConsumedExpr(cast<Expr>(S)))
621 return PM.getParentIgnoreParens(S);
622
623 const Stmt *Parent = PM.getParentIgnoreParens(S);
624 if (!Parent)
625 return nullptr;
626
627 switch (Parent->getStmtClass()) {
628 case Stmt::ForStmtClass:
629 case Stmt::DoStmtClass:
630 case Stmt::WhileStmtClass:
631 case Stmt::ObjCForCollectionStmtClass:
632 case Stmt::CXXForRangeStmtClass:
633 return Parent;
634 default:
635 break;
636 }
637
638 return nullptr;
639}
640
641static PathDiagnosticLocation
642getEnclosingStmtLocation(const Stmt *S, const LocationContext *LC,
643 bool allowNestedContexts = false) {
644 if (!S)
645 return {};
646
647 const SourceManager &SMgr = LC->getDecl()->getASTContext().getSourceManager();
648
649 while (const Stmt *Parent = getEnclosingParent(S, LC->getParentMap())) {
650 switch (Parent->getStmtClass()) {
651 case Stmt::BinaryOperatorClass: {
652 const auto *B = cast<BinaryOperator>(Parent);
653 if (B->isLogicalOp())
654 return PathDiagnosticLocation(allowNestedContexts ? B : S, SMgr, LC);
655 break;
656 }
657 case Stmt::CompoundStmtClass:
658 case Stmt::StmtExprClass:
659 return PathDiagnosticLocation(S, SMgr, LC);
660 case Stmt::ChooseExprClass:
661 // Similar to '?' if we are referring to condition, just have the edge
662 // point to the entire choose expression.
663 if (allowNestedContexts || cast<ChooseExpr>(Parent)->getCond() == S)
664 return PathDiagnosticLocation(Parent, SMgr, LC);
665 else
666 return PathDiagnosticLocation(S, SMgr, LC);
667 case Stmt::BinaryConditionalOperatorClass:
668 case Stmt::ConditionalOperatorClass:
669 // For '?', if we are referring to condition, just have the edge point
670 // to the entire '?' expression.
671 if (allowNestedContexts ||
672 cast<AbstractConditionalOperator>(Parent)->getCond() == S)
673 return PathDiagnosticLocation(Parent, SMgr, LC);
674 else
675 return PathDiagnosticLocation(S, SMgr, LC);
676 case Stmt::CXXForRangeStmtClass:
677 if (cast<CXXForRangeStmt>(Parent)->getBody() == S)
678 return PathDiagnosticLocation(S, SMgr, LC);
679 break;
680 case Stmt::DoStmtClass:
681 return PathDiagnosticLocation(S, SMgr, LC);
682 case Stmt::ForStmtClass:
683 if (cast<ForStmt>(Parent)->getBody() == S)
684 return PathDiagnosticLocation(S, SMgr, LC);
685 break;
686 case Stmt::IfStmtClass:
687 if (cast<IfStmt>(Parent)->getCond() != S)
688 return PathDiagnosticLocation(S, SMgr, LC);
689 break;
690 case Stmt::ObjCForCollectionStmtClass:
691 if (cast<ObjCForCollectionStmt>(Parent)->getBody() == S)
692 return PathDiagnosticLocation(S, SMgr, LC);
693 break;
694 case Stmt::WhileStmtClass:
695 if (cast<WhileStmt>(Parent)->getCond() != S)
696 return PathDiagnosticLocation(S, SMgr, LC);
697 break;
698 default:
699 break;
700 }
701
702 S = Parent;
703 }
704
705 assert(S && "Cannot have null Stmt for PathDiagnosticLocation");
706
707 return PathDiagnosticLocation(S, SMgr, LC);
708}
709
710//===----------------------------------------------------------------------===//
711// "Minimal" path diagnostic generation algorithm.
712//===----------------------------------------------------------------------===//
713
714/// If the piece contains a special message, add it to all the call pieces on
715/// the active stack. For example, my_malloc allocated memory, so MallocChecker
716/// will construct an event at the call to malloc(), and add a stack hint that
717/// an allocated memory was returned. We'll use this hint to construct a message
718/// when returning from the call to my_malloc
719///
720/// void *my_malloc() { return malloc(sizeof(int)); }
721/// void fishy() {
722/// void *ptr = my_malloc(); // returned allocated memory
723/// } // leak
724void PathDiagnosticBuilder::updateStackPiecesWithMessage(
725 PathDiagnosticPieceRef P, const CallWithEntryStack &CallStack) const {
726 if (R->hasCallStackHint(P))
727 for (const auto &I : CallStack) {
728 PathDiagnosticCallPiece *CP = I.first;
729 const ExplodedNode *N = I.second;
730 std::string stackMsg = R->getCallStackMessage(P, N);
731
732 // The last message on the path to final bug is the most important
733 // one. Since we traverse the path backwards, do not add the message
734 // if one has been previously added.
735 if (!CP->hasCallStackMessage())
736 CP->setCallStackMessage(stackMsg);
737 }
738}
739
740static void CompactMacroExpandedPieces(PathPieces &path,
741 const SourceManager& SM);
742
743PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForSwitchOP(
744 const PathDiagnosticConstruct &C, const CFGBlock *Dst,
745 PathDiagnosticLocation &Start) const {
746
747 const SourceManager &SM = getSourceManager();
748 // Figure out what case arm we took.
749 std::string sbuf;
750 llvm::raw_string_ostream os(sbuf);
751 PathDiagnosticLocation End;
752
753 if (const Stmt *S = Dst->getLabel()) {
754 End = PathDiagnosticLocation(S, SM, C.getCurrLocationContext());
755
756 switch (S->getStmtClass()) {
757 default:
758 os << "No cases match in the switch statement. "
759 "Control jumps to line "
760 << End.asLocation().getExpansionLineNumber();
761 break;
762 case Stmt::DefaultStmtClass:
763 os << "Control jumps to the 'default' case at line "
764 << End.asLocation().getExpansionLineNumber();
765 break;
766
767 case Stmt::CaseStmtClass: {
768 os << "Control jumps to 'case ";
769 const auto *Case = cast<CaseStmt>(S);
770 const Expr *LHS = Case->getLHS()->IgnoreParenCasts();
771
772 // Determine if it is an enum.
773 bool GetRawInt = true;
774
775 if (const auto *DR = dyn_cast<DeclRefExpr>(LHS)) {
776 // FIXME: Maybe this should be an assertion. Are there cases
777 // were it is not an EnumConstantDecl?
778 const auto *D = dyn_cast<EnumConstantDecl>(DR->getDecl());
779
780 if (D) {
781 GetRawInt = false;
782 os << *D;
783 }
784 }
785
786 if (GetRawInt)
787 os << LHS->EvaluateKnownConstInt(getASTContext());
788
789 os << ":' at line " << End.asLocation().getExpansionLineNumber();
790 break;
791 }
792 }
793 } else {
794 os << "'Default' branch taken. ";
795 End = ExecutionContinues(os, C);
796 }
797 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End,
798 os.str());
799}
800
801PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForGotoOP(
802 const PathDiagnosticConstruct &C, const Stmt *S,
803 PathDiagnosticLocation &Start) const {
804 std::string sbuf;
805 llvm::raw_string_ostream os(sbuf);
806 const PathDiagnosticLocation &End =
807 getEnclosingStmtLocation(S, C.getCurrLocationContext());
808 os << "Control jumps to line " << End.asLocation().getExpansionLineNumber();
809 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str());
810}
811
812PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForBinaryOP(
813 const PathDiagnosticConstruct &C, const Stmt *T, const CFGBlock *Src,
814 const CFGBlock *Dst) const {
815
816 const SourceManager &SM = getSourceManager();
817
818 const auto *B = cast<BinaryOperator>(T);
819 std::string sbuf;
820 llvm::raw_string_ostream os(sbuf);
821 os << "Left side of '";
822 PathDiagnosticLocation Start, End;
823
824 if (B->getOpcode() == BO_LAnd) {
825 os << "&&"
826 << "' is ";
827
828 if (*(Src->succ_begin() + 1) == Dst) {
829 os << "false";
830 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext());
831 Start =
832 PathDiagnosticLocation::createOperatorLoc(B, SM);
833 } else {
834 os << "true";
835 Start =
836 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext());
837 End = ExecutionContinues(C);
838 }
839 } else {
840 assert(B->getOpcode() == BO_LOr);
841 os << "||"
842 << "' is ";
843
844 if (*(Src->succ_begin() + 1) == Dst) {
845 os << "false";
846 Start =
847 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext());
848 End = ExecutionContinues(C);
849 } else {
850 os << "true";
851 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext());
852 Start =
853 PathDiagnosticLocation::createOperatorLoc(B, SM);
854 }
855 }
856 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End,
857 os.str());
858}
859
860void PathDiagnosticBuilder::generateMinimalDiagForBlockEdge(
861 PathDiagnosticConstruct &C, BlockEdge BE) const {
862 const SourceManager &SM = getSourceManager();
863 const LocationContext *LC = C.getCurrLocationContext();
864 const CFGBlock *Src = BE.getSrc();
865 const CFGBlock *Dst = BE.getDst();
866 const Stmt *T = Src->getTerminatorStmt();
867 if (!T)
868 return;
869
870 auto Start = PathDiagnosticLocation::createBegin(T, SM, LC);
871 switch (T->getStmtClass()) {
872 default:
873 break;
874
875 case Stmt::GotoStmtClass:
876 case Stmt::IndirectGotoStmtClass: {
877 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics())
878 C.getActivePath().push_front(generateDiagForGotoOP(C, S, Start));
879 break;
880 }
881
882 case Stmt::SwitchStmtClass: {
883 C.getActivePath().push_front(generateDiagForSwitchOP(C, Dst, Start));
884 break;
885 }
886
887 case Stmt::BreakStmtClass:
888 case Stmt::ContinueStmtClass: {
889 std::string sbuf;
890 llvm::raw_string_ostream os(sbuf);
891 PathDiagnosticLocation End = ExecutionContinues(os, C);
892 C.getActivePath().push_front(
893 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str()));
894 break;
895 }
896
897 // Determine control-flow for ternary '?'.
898 case Stmt::BinaryConditionalOperatorClass:
899 case Stmt::ConditionalOperatorClass: {
900 std::string sbuf;
901 llvm::raw_string_ostream os(sbuf);
902 os << "'?' condition is ";
903
904 if (*(Src->succ_begin() + 1) == Dst)
905 os << "false";
906 else
907 os << "true";
908
909 PathDiagnosticLocation End = ExecutionContinues(C);
910
911 if (const Stmt *S = End.asStmt())
912 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
913
914 C.getActivePath().push_front(
915 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str()));
916 break;
917 }
918
919 // Determine control-flow for short-circuited '&&' and '||'.
920 case Stmt::BinaryOperatorClass: {
921 if (!C.supportsLogicalOpControlFlow())
922 break;
923
924 C.getActivePath().push_front(generateDiagForBinaryOP(C, T, Src, Dst));
925 break;
926 }
927
928 case Stmt::DoStmtClass:
929 if (*(Src->succ_begin()) == Dst) {
930 std::string sbuf;
931 llvm::raw_string_ostream os(sbuf);
932
933 os << "Loop condition is true. ";
934 PathDiagnosticLocation End = ExecutionContinues(os, C);
935
936 if (const Stmt *S = End.asStmt())
937 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
938
939 C.getActivePath().push_front(
940 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End,
941 os.str()));
942 } else {
943 PathDiagnosticLocation End = ExecutionContinues(C);
944
945 if (const Stmt *S = End.asStmt())
946 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
947
948 C.getActivePath().push_front(
949 std::make_shared<PathDiagnosticControlFlowPiece>(
950 Start, End, "Loop condition is false. Exiting loop"));
951 }
952 break;
953
954 case Stmt::WhileStmtClass:
955 case Stmt::ForStmtClass:
956 if (*(Src->succ_begin() + 1) == Dst) {
957 std::string sbuf;
958 llvm::raw_string_ostream os(sbuf);
959
960 os << "Loop condition is false. ";
961 PathDiagnosticLocation End = ExecutionContinues(os, C);
962 if (const Stmt *S = End.asStmt())
963 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
964
965 C.getActivePath().push_front(
966 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End,
967 os.str()));
968 } else {
969 PathDiagnosticLocation End = ExecutionContinues(C);
970 if (const Stmt *S = End.asStmt())
971 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
972
973 C.getActivePath().push_front(
974 std::make_shared<PathDiagnosticControlFlowPiece>(
975 Start, End, "Loop condition is true. Entering loop body"));
976 }
977
978 break;
979
980 case Stmt::IfStmtClass: {
981 PathDiagnosticLocation End = ExecutionContinues(C);
982
983 if (const Stmt *S = End.asStmt())
984 End = getEnclosingStmtLocation(S, C.getCurrLocationContext());
985
986 if (*(Src->succ_begin() + 1) == Dst)
987 C.getActivePath().push_front(
988 std::make_shared<PathDiagnosticControlFlowPiece>(
989 Start, End, "Taking false branch"));
990 else
991 C.getActivePath().push_front(
992 std::make_shared<PathDiagnosticControlFlowPiece>(
993 Start, End, "Taking true branch"));
994
995 break;
996 }
997 }
998}
999
1000//===----------------------------------------------------------------------===//
1001// Functions for determining if a loop was executed 0 times.
1002//===----------------------------------------------------------------------===//
1003
1004static bool isLoop(const Stmt *Term) {
1005 switch (Term->getStmtClass()) {
1006 case Stmt::ForStmtClass:
1007 case Stmt::WhileStmtClass:
1008 case Stmt::ObjCForCollectionStmtClass:
1009 case Stmt::CXXForRangeStmtClass:
1010 return true;
1011 default:
1012 // Note that we intentionally do not include do..while here.
1013 return false;
1014 }
1015}
1016
1017static bool isJumpToFalseBranch(const BlockEdge *BE) {
1018 const CFGBlock *Src = BE->getSrc();
1019 assert(Src->succ_size() == 2);
1020 return (*(Src->succ_begin()+1) == BE->getDst());
1021}
1022
1023static bool isContainedByStmt(const ParentMap &PM, const Stmt *S,
1024 const Stmt *SubS) {
1025 while (SubS) {
1026 if (SubS == S)
1027 return true;
1028 SubS = PM.getParent(SubS);
1029 }
1030 return false;
1031}
1032
1033static const Stmt *getStmtBeforeCond(const ParentMap &PM, const Stmt *Term,
1034 const ExplodedNode *N) {
1035 while (N) {
1036 Optional<StmtPoint> SP = N->getLocation().getAs<StmtPoint>();
1037 if (SP) {
1038 const Stmt *S = SP->getStmt();
1039 if (!isContainedByStmt(PM, Term, S))
1040 return S;
1041 }
1042 N = N->getFirstPred();
1043 }
1044 return nullptr;
1045}
1046
1047static bool isInLoopBody(const ParentMap &PM, const Stmt *S, const Stmt *Term) {
1048 const Stmt *LoopBody = nullptr;
1049 switch (Term->getStmtClass()) {
1050 case Stmt::CXXForRangeStmtClass: {
1051 const auto *FR = cast<CXXForRangeStmt>(Term);
1052 if (isContainedByStmt(PM, FR->getInc(), S))
1053 return true;
1054 if (isContainedByStmt(PM, FR->getLoopVarStmt(), S))
1055 return true;
1056 LoopBody = FR->getBody();
1057 break;
1058 }
1059 case Stmt::ForStmtClass: {
1060 const auto *FS = cast<ForStmt>(Term);
1061 if (isContainedByStmt(PM, FS->getInc(), S))
1062 return true;
1063 LoopBody = FS->getBody();
1064 break;
1065 }
1066 case Stmt::ObjCForCollectionStmtClass: {
1067 const auto *FC = cast<ObjCForCollectionStmt>(Term);
1068 LoopBody = FC->getBody();
1069 break;
1070 }
1071 case Stmt::WhileStmtClass:
1072 LoopBody = cast<WhileStmt>(Term)->getBody();
1073 break;
1074 default:
1075 return false;
1076 }
1077 return isContainedByStmt(PM, LoopBody, S);
1078}
1079
1080/// Adds a sanitized control-flow diagnostic edge to a path.
1081static void addEdgeToPath(PathPieces &path,
1082 PathDiagnosticLocation &PrevLoc,
1083 PathDiagnosticLocation NewLoc) {
1084 if (!NewLoc.isValid())
1085 return;
1086
1087 SourceLocation NewLocL = NewLoc.asLocation();
1088 if (NewLocL.isInvalid())
1089 return;
1090
1091 if (!PrevLoc.isValid() || !PrevLoc.asLocation().isValid()) {
1092 PrevLoc = NewLoc;
1093 return;
1094 }
1095
1096 // Ignore self-edges, which occur when there are multiple nodes at the same
1097 // statement.
1098 if (NewLoc.asStmt() && NewLoc.asStmt() == PrevLoc.asStmt())
1099 return;
1100
1101 path.push_front(
1102 std::make_shared<PathDiagnosticControlFlowPiece>(NewLoc, PrevLoc));
1103 PrevLoc = NewLoc;
1104}
1105
1106/// A customized wrapper for CFGBlock::getTerminatorCondition()
1107/// which returns the element for ObjCForCollectionStmts.
1108static const Stmt *getTerminatorCondition(const CFGBlock *B) {
1109 const Stmt *S = B->getTerminatorCondition();
1110 if (const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(S))
1111 return FS->getElement();
1112 return S;
1113}
1114
1115constexpr llvm::StringLiteral StrEnteringLoop = "Entering loop body";
1116constexpr llvm::StringLiteral StrLoopBodyZero = "Loop body executed 0 times";
1117constexpr llvm::StringLiteral StrLoopRangeEmpty =
1118 "Loop body skipped when range is empty";
1119constexpr llvm::StringLiteral StrLoopCollectionEmpty =
1120 "Loop body skipped when collection is empty";
1121
1122static std::unique_ptr<FilesToLineNumsMap>
1123findExecutedLines(const SourceManager &SM, const ExplodedNode *N);
1124
1125void PathDiagnosticBuilder::generatePathDiagnosticsForNode(
1126 PathDiagnosticConstruct &C, PathDiagnosticLocation &PrevLoc) const {
1127 ProgramPoint P = C.getCurrentNode()->getLocation();
1128 const SourceManager &SM = getSourceManager();
1129
1130 // Have we encountered an entrance to a call? It may be
1131 // the case that we have not encountered a matching
1132 // call exit before this point. This means that the path
1133 // terminated within the call itself.
1134 if (auto CE = P.getAs<CallEnter>()) {
1135
1136 if (C.shouldAddPathEdges()) {
1137 // Add an edge to the start of the function.
1138 const StackFrameContext *CalleeLC = CE->getCalleeContext();
1139 const Decl *D = CalleeLC->getDecl();
1140 // Add the edge only when the callee has body. We jump to the beginning
1141 // of the *declaration*, however we expect it to be followed by the
1142 // body. This isn't the case for autosynthesized property accessors in
1143 // Objective-C. No need for a similar extra check for CallExit points
1144 // because the exit edge comes from a statement (i.e. return),
1145 // not from declaration.
1146 if (D->hasBody())
1147 addEdgeToPath(C.getActivePath(), PrevLoc,
1148 PathDiagnosticLocation::createBegin(D, SM));
1149 }
1150
1151 // Did we visit an entire call?
1152 bool VisitedEntireCall = C.PD->isWithinCall();
1153 C.PD->popActivePath();
1154
1155 PathDiagnosticCallPiece *Call;
1156 if (VisitedEntireCall) {
1157 Call = cast<PathDiagnosticCallPiece>(C.getActivePath().front().get());
1158 } else {
1159 // The path terminated within a nested location context, create a new
1160 // call piece to encapsulate the rest of the path pieces.
1161 const Decl *Caller = CE->getLocationContext()->getDecl();
1162 Call = PathDiagnosticCallPiece::construct(C.getActivePath(), Caller);
1163 assert(C.getActivePath().size() == 1 &&
1164 C.getActivePath().front().get() == Call);
1165
1166 // Since we just transferred the path over to the call piece, reset the
1167 // mapping of the active path to the current location context.
1168 assert(C.isInLocCtxMap(&C.getActivePath()) &&
1169 "When we ascend to a previously unvisited call, the active path's "
1170 "address shouldn't change, but rather should be compacted into "
1171 "a single CallEvent!");
1172 C.updateLocCtxMap(&C.getActivePath(), C.getCurrLocationContext());
1173
1174 // Record the location context mapping for the path within the call.
1175 assert(!C.isInLocCtxMap(&Call->path) &&
1176 "When we ascend to a previously unvisited call, this must be the "
1177 "first time we encounter the caller context!");
1178 C.updateLocCtxMap(&Call->path, CE->getCalleeContext());
1179 }
1180 Call->setCallee(*CE, SM);
1181
1182 // Update the previous location in the active path.
1183 PrevLoc = Call->getLocation();
1184
1185 if (!C.CallStack.empty()) {
1186 assert(C.CallStack.back().first == Call);
1187 C.CallStack.pop_back();
1188 }
1189 return;
1190 }
1191
1192 assert(C.getCurrLocationContext() == C.getLocationContextForActivePath() &&
1193 "The current position in the bug path is out of sync with the "
1194 "location context associated with the active path!");
1195
1196 // Have we encountered an exit from a function call?
1197 if (Optional<CallExitEnd> CE = P.getAs<CallExitEnd>()) {
1198
1199 // We are descending into a call (backwards). Construct
1200 // a new call piece to contain the path pieces for that call.
1201 auto Call = PathDiagnosticCallPiece::construct(*CE, SM);
1202 // Record the mapping from call piece to LocationContext.
1203 assert(!C.isInLocCtxMap(&Call->path) &&
1204 "We just entered a call, this must've been the first time we "
1205 "encounter its context!");
1206 C.updateLocCtxMap(&Call->path, CE->getCalleeContext());
1207
1208 if (C.shouldAddPathEdges()) {
1209 // Add the edge to the return site.
1210 addEdgeToPath(C.getActivePath(), PrevLoc, Call->callReturn);
1211 PrevLoc.invalidate();
1212 }
1213
1214 auto *P = Call.get();
1215 C.getActivePath().push_front(std::move(Call));
1216
1217 // Make the contents of the call the active path for now.
1218 C.PD->pushActivePath(&P->path);
1219 C.CallStack.push_back(CallWithEntry(P, C.getCurrentNode()));
1220 return;
1221 }
1222
1223 if (auto PS = P.getAs<PostStmt>()) {
1224 if (!C.shouldAddPathEdges())
1225 return;
1226
1227 // Add an edge. If this is an ObjCForCollectionStmt do
1228 // not add an edge here as it appears in the CFG both
1229 // as a terminator and as a terminator condition.
1230 if (!isa<ObjCForCollectionStmt>(PS->getStmt())) {
1231 PathDiagnosticLocation L =
1232 PathDiagnosticLocation(PS->getStmt(), SM, C.getCurrLocationContext());
1233 addEdgeToPath(C.getActivePath(), PrevLoc, L);
1234 }
1235
1236 } else if (auto BE = P.getAs<BlockEdge>()) {
1237
1238 if (C.shouldAddControlNotes()) {
1239 generateMinimalDiagForBlockEdge(C, *BE);
1240 }
1241
1242 if (!C.shouldAddPathEdges()) {
1243 return;
1244 }
1245
1246 // Are we jumping to the head of a loop? Add a special diagnostic.
1247 if (const Stmt *Loop = BE->getSrc()->getLoopTarget()) {
1248 PathDiagnosticLocation L(Loop, SM, C.getCurrLocationContext());
1249 const Stmt *Body = nullptr;
1250
1251 if (const auto *FS = dyn_cast<ForStmt>(Loop))
1252 Body = FS->getBody();
1253 else if (const auto *WS = dyn_cast<WhileStmt>(Loop))
1254 Body = WS->getBody();
1255 else if (const auto *OFS = dyn_cast<ObjCForCollectionStmt>(Loop)) {
1256 Body = OFS->getBody();
1257 } else if (const auto *FRS = dyn_cast<CXXForRangeStmt>(Loop)) {
1258 Body = FRS->getBody();
1259 }
1260 // do-while statements are explicitly excluded here
1261
1262 auto p = std::make_shared<PathDiagnosticEventPiece>(
1263 L, "Looping back to the head of the loop");
1264 p->setPrunable(true);
1265
1266 addEdgeToPath(C.getActivePath(), PrevLoc, p->getLocation());
1267 // We might've added a very similar control node already
1268 if (!C.shouldAddControlNotes()) {
1269 C.getActivePath().push_front(std::move(p));
1270 }
1271
1272 if (const auto *CS = dyn_cast_or_null<CompoundStmt>(Body)) {
1273 addEdgeToPath(C.getActivePath(), PrevLoc,
1274 PathDiagnosticLocation::createEndBrace(CS, SM));
1275 }
1276 }
1277
1278 const CFGBlock *BSrc = BE->getSrc();
1279 const ParentMap &PM = C.getParentMap();
1280
1281 if (const Stmt *Term = BSrc->getTerminatorStmt()) {
1282 // Are we jumping past the loop body without ever executing the
1283 // loop (because the condition was false)?
1284 if (isLoop(Term)) {
1285 const Stmt *TermCond = getTerminatorCondition(BSrc);
1286 bool IsInLoopBody = isInLoopBody(
1287 PM, getStmtBeforeCond(PM, TermCond, C.getCurrentNode()), Term);
1288
1289 StringRef str;
1290
1291 if (isJumpToFalseBranch(&*BE)) {
1292 if (!IsInLoopBody) {
1293 if (isa<ObjCForCollectionStmt>(Term)) {
1294 str = StrLoopCollectionEmpty;
1295 } else if (isa<CXXForRangeStmt>(Term)) {
1296 str = StrLoopRangeEmpty;
1297 } else {
1298 str = StrLoopBodyZero;
1299 }
1300 }
1301 } else {
1302 str = StrEnteringLoop;
1303 }
1304
1305 if (!str.empty()) {
1306 PathDiagnosticLocation L(TermCond ? TermCond : Term, SM,
1307 C.getCurrLocationContext());
1308 auto PE = std::make_shared<PathDiagnosticEventPiece>(L, str);
1309 PE->setPrunable(true);
1310 addEdgeToPath(C.getActivePath(), PrevLoc, PE->getLocation());
1311
1312 // We might've added a very similar control node already
1313 if (!C.shouldAddControlNotes()) {
1314 C.getActivePath().push_front(std::move(PE));
1315 }
1316 }
1317 } else if (isa<BreakStmt, ContinueStmt, GotoStmt>(Term)) {
1318 PathDiagnosticLocation L(Term, SM, C.getCurrLocationContext());
1319 addEdgeToPath(C.getActivePath(), PrevLoc, L);
1320 }
1321 }
1322 }
1323}
1324
1325static std::unique_ptr<PathDiagnostic>
1326generateDiagnosticForBasicReport(const BasicBugReport *R) {
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 std::make_unique<FilesToLineNumsMap>());
1333}
1334
1335static std::unique_ptr<PathDiagnostic>
1336generateEmptyDiagnosticForReport(const PathSensitiveBugReport *R,
1337 const SourceManager &SM) {
1338 const BugType &BT = R->getBugType();
1339 return std::make_unique<PathDiagnostic>(
1340 BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(),
1341 R->getDescription(), R->getShortDescription(/*UseFallback=*/false),
1342 BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(),
1343 findExecutedLines(SM, R->getErrorNode()));
1344}
1345
1346static const Stmt *getStmtParent(const Stmt *S, const ParentMap &PM) {
1347 if (!S)
1348 return nullptr;
1349
1350 while (true) {
1351 S = PM.getParentIgnoreParens(S);
1352
1353 if (!S)
1354 break;
1355
1356 if (isa<FullExpr, CXXBindTemporaryExpr, SubstNonTypeTemplateParmExpr>(S))
1357 continue;
1358
1359 break;
1360 }
1361
1362 return S;
1363}
1364
1365static bool isConditionForTerminator(const Stmt *S, const Stmt *Cond) {
1366 switch (S->getStmtClass()) {
1367 case Stmt::BinaryOperatorClass: {
1368 const auto *BO = cast<BinaryOperator>(S);
1369 if (!BO->isLogicalOp())
1370 return false;
1371 return BO->getLHS() == Cond || BO->getRHS() == Cond;
1372 }
1373 case Stmt::IfStmtClass:
1374 return cast<IfStmt>(S)->getCond() == Cond;
1375 case Stmt::ForStmtClass:
1376 return cast<ForStmt>(S)->getCond() == Cond;
1377 case Stmt::WhileStmtClass:
1378 return cast<WhileStmt>(S)->getCond() == Cond;
1379 case Stmt::DoStmtClass:
1380 return cast<DoStmt>(S)->getCond() == Cond;
1381 case Stmt::ChooseExprClass:
1382 return cast<ChooseExpr>(S)->getCond() == Cond;
1383 case Stmt::IndirectGotoStmtClass:
1384 return cast<IndirectGotoStmt>(S)->getTarget() == Cond;
1385 case Stmt::SwitchStmtClass:
1386 return cast<SwitchStmt>(S)->getCond() == Cond;
1387 case Stmt::BinaryConditionalOperatorClass:
1388 return cast<BinaryConditionalOperator>(S)->getCond() == Cond;
1389 case Stmt::ConditionalOperatorClass: {
1390 const auto *CO = cast<ConditionalOperator>(S);
1391 return CO->getCond() == Cond ||
1392 CO->getLHS() == Cond ||
1393 CO->getRHS() == Cond;
1394 }
1395 case Stmt::ObjCForCollectionStmtClass:
1396 return cast<ObjCForCollectionStmt>(S)->getElement() == Cond;
1397 case Stmt::CXXForRangeStmtClass: {
1398 const auto *FRS = cast<CXXForRangeStmt>(S);
1399 return FRS->getCond() == Cond || FRS->getRangeInit() == Cond;
1400 }
1401 default:
1402 return false;
1403 }
1404}
1405
1406static bool isIncrementOrInitInForLoop(const Stmt *S, const Stmt *FL) {
1407 if (const auto *FS = dyn_cast<ForStmt>(FL))
1408 return FS->getInc() == S || FS->getInit() == S;
1409 if (const auto *FRS = dyn_cast<CXXForRangeStmt>(FL))
1410 return FRS->getInc() == S || FRS->getRangeStmt() == S ||
1411 FRS->getLoopVarStmt() || FRS->getRangeInit() == S;
1412 return false;
1413}
1414
1415using OptimizedCallsSet = llvm::DenseSet<const PathDiagnosticCallPiece *>;
1416
1417/// Adds synthetic edges from top-level statements to their subexpressions.
1418///
1419/// This avoids a "swoosh" effect, where an edge from a top-level statement A
1420/// points to a sub-expression B.1 that's not at the start of B. In these cases,
1421/// we'd like to see an edge from A to B, then another one from B to B.1.
1422static void addContextEdges(PathPieces &pieces, const LocationContext *LC) {
1423 const ParentMap &PM = LC->getParentMap();
1424 PathPieces::iterator Prev = pieces.end();
1425 for (PathPieces::iterator I = pieces.begin(), E = Prev; I != E;
1426 Prev = I, ++I) {
1427 auto *Piece = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1428
1429 if (!Piece)
1430 continue;
1431
1432 PathDiagnosticLocation SrcLoc = Piece->getStartLocation();
1433 SmallVector<PathDiagnosticLocation, 4> SrcContexts;
1434
1435 PathDiagnosticLocation NextSrcContext = SrcLoc;
1436 const Stmt *InnerStmt = nullptr;
1437 while (NextSrcContext.isValid() && NextSrcContext.asStmt() != InnerStmt) {
1438 SrcContexts.push_back(NextSrcContext);
1439 InnerStmt = NextSrcContext.asStmt();
1440 NextSrcContext = getEnclosingStmtLocation(InnerStmt, LC,
1441 /*allowNested=*/true);
1442 }
1443
1444 // Repeatedly split the edge as necessary.
1445 // This is important for nested logical expressions (||, &&, ?:) where we
1446 // want to show all the levels of context.
1447 while (true) {
1448 const Stmt *Dst = Piece->getEndLocation().getStmtOrNull();
1449
1450 // We are looking at an edge. Is the destination within a larger
1451 // expression?
1452 PathDiagnosticLocation DstContext =
1453 getEnclosingStmtLocation(Dst, LC, /*allowNested=*/true);
1454 if (!DstContext.isValid() || DstContext.asStmt() == Dst)
1455 break;
1456
1457 // If the source is in the same context, we're already good.
1458 if (llvm::is_contained(SrcContexts, DstContext))
1459 break;
1460
1461 // Update the subexpression node to point to the context edge.
1462 Piece->setStartLocation(DstContext);
1463
1464 // Try to extend the previous edge if it's at the same level as the source
1465 // context.
1466 if (Prev != E) {
1467 auto *PrevPiece = dyn_cast<PathDiagnosticControlFlowPiece>(Prev->get());
1468
1469 if (PrevPiece) {
1470 if (const Stmt *PrevSrc =
1471 PrevPiece->getStartLocation().getStmtOrNull()) {
1472 const Stmt *PrevSrcParent = getStmtParent(PrevSrc, PM);
1473 if (PrevSrcParent ==
1474 getStmtParent(DstContext.getStmtOrNull(), PM)) {
1475 PrevPiece->setEndLocation(DstContext);
1476 break;
1477 }
1478 }
1479 }
1480 }
1481
1482 // Otherwise, split the current edge into a context edge and a
1483 // subexpression edge. Note that the context statement may itself have
1484 // context.
1485 auto P =
1486 std::make_shared<PathDiagnosticControlFlowPiece>(SrcLoc, DstContext);
1487 Piece = P.get();
1488 I = pieces.insert(I, std::move(P));
1489 }
1490 }
1491}
1492
1493/// Move edges from a branch condition to a branch target
1494/// when the condition is simple.
1495///
1496/// This restructures some of the work of addContextEdges. That function
1497/// creates edges this may destroy, but they work together to create a more
1498/// aesthetically set of edges around branches. After the call to
1499/// addContextEdges, we may have (1) an edge to the branch, (2) an edge from
1500/// the branch to the branch condition, and (3) an edge from the branch
1501/// condition to the branch target. We keep (1), but may wish to remove (2)
1502/// and move the source of (3) to the branch if the branch condition is simple.
1503static void simplifySimpleBranches(PathPieces &pieces) {
1504 for (PathPieces::iterator I = pieces.begin(), E = pieces.end(); I != E; ++I) {
1505 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1506
1507 if (!PieceI)
1508 continue;
1509
1510 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1511 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull();
1512
1513 if (!s1Start || !s1End)
1514 continue;
1515
1516 PathPieces::iterator NextI = I; ++NextI;
1517 if (NextI == E)
1518 break;
1519
1520 PathDiagnosticControlFlowPiece *PieceNextI = nullptr;
1521
1522 while (true) {
1523 if (NextI == E)
1524 break;
1525
1526 const auto *EV = dyn_cast<PathDiagnosticEventPiece>(NextI->get());
1527 if (EV) {
1528 StringRef S = EV->getString();
1529 if (S == StrEnteringLoop || S == StrLoopBodyZero ||
1530 S == StrLoopCollectionEmpty || S == StrLoopRangeEmpty) {
1531 ++NextI;
1532 continue;
1533 }
1534 break;
1535 }
1536
1537 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1538 break;
1539 }
1540
1541 if (!PieceNextI)
1542 continue;
1543
1544 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1545 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull();
1546
1547 if (!s2Start || !s2End || s1End != s2Start)
1548 continue;
1549
1550 // We only perform this transformation for specific branch kinds.
1551 // We don't want to do this for do..while, for example.
1552 if (!isa<ForStmt, WhileStmt, IfStmt, ObjCForCollectionStmt,
1553 CXXForRangeStmt>(s1Start))
1554 continue;
1555
1556 // Is s1End the branch condition?
1557 if (!isConditionForTerminator(s1Start, s1End))
1558 continue;
1559
1560 // Perform the hoisting by eliminating (2) and changing the start
1561 // location of (3).
1562 PieceNextI->setStartLocation(PieceI->getStartLocation());
1563 I = pieces.erase(I);
1564 }
1565}
1566
1567/// Returns the number of bytes in the given (character-based) SourceRange.
1568///
1569/// If the locations in the range are not on the same line, returns None.
1570///
1571/// Note that this does not do a precise user-visible character or column count.
1572static Optional<size_t> getLengthOnSingleLine(const SourceManager &SM,
1573 SourceRange Range) {
1574 SourceRange ExpansionRange(SM.getExpansionLoc(Range.getBegin()),
1575 SM.getExpansionRange(Range.getEnd()).getEnd());
1576
1577 FileID FID = SM.getFileID(ExpansionRange.getBegin());
1578 if (FID != SM.getFileID(ExpansionRange.getEnd()))
1579 return None;
1580
1581 Optional<MemoryBufferRef> Buffer = SM.getBufferOrNone(FID);
1582 if (!Buffer)
1583 return None;
1584
1585 unsigned BeginOffset = SM.getFileOffset(ExpansionRange.getBegin());
1586 unsigned EndOffset = SM.getFileOffset(ExpansionRange.getEnd());
1587 StringRef Snippet = Buffer->getBuffer().slice(BeginOffset, EndOffset);
1588
1589 // We're searching the raw bytes of the buffer here, which might include
1590 // escaped newlines and such. That's okay; we're trying to decide whether the
1591 // SourceRange is covering a large or small amount of space in the user's
1592 // editor.
1593 if (Snippet.find_first_of("\r\n") != StringRef::npos)
1594 return None;
1595
1596 // This isn't Unicode-aware, but it doesn't need to be.
1597 return Snippet.size();
1598}
1599
1600/// \sa getLengthOnSingleLine(SourceManager, SourceRange)
1601static Optional<size_t> getLengthOnSingleLine(const SourceManager &SM,
1602 const Stmt *S) {
1603 return getLengthOnSingleLine(SM, S->getSourceRange());
1604}
1605
1606/// Eliminate two-edge cycles created by addContextEdges().
1607///
1608/// Once all the context edges are in place, there are plenty of cases where
1609/// there's a single edge from a top-level statement to a subexpression,
1610/// followed by a single path note, and then a reverse edge to get back out to
1611/// the top level. If the statement is simple enough, the subexpression edges
1612/// just add noise and make it harder to understand what's going on.
1613///
1614/// This function only removes edges in pairs, because removing only one edge
1615/// might leave other edges dangling.
1616///
1617/// This will not remove edges in more complicated situations:
1618/// - if there is more than one "hop" leading to or from a subexpression.
1619/// - if there is an inlined call between the edges instead of a single event.
1620/// - if the whole statement is large enough that having subexpression arrows
1621/// might be helpful.
1622static void removeContextCycles(PathPieces &Path, const SourceManager &SM) {
1623 for (PathPieces::iterator I = Path.begin(), E = Path.end(); I != E; ) {
1624 // Pattern match the current piece and its successor.
1625 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1626
1627 if (!PieceI) {
1628 ++I;
1629 continue;
1630 }
1631
1632 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1633 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull();
1634
1635 PathPieces::iterator NextI = I; ++NextI;
1636 if (NextI == E)
1637 break;
1638
1639 const auto *PieceNextI =
1640 dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1641
1642 if (!PieceNextI) {
1643 if (isa<PathDiagnosticEventPiece>(NextI->get())) {
1644 ++NextI;
1645 if (NextI == E)
1646 break;
1647 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1648 }
1649
1650 if (!PieceNextI) {
1651 ++I;
1652 continue;
1653 }
1654 }
1655
1656 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1657 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull();
1658
1659 if (s1Start && s2Start && s1Start == s2End && s2Start == s1End) {
1660 const size_t MAX_SHORT_LINE_LENGTH = 80;
1661 Optional<size_t> s1Length = getLengthOnSingleLine(SM, s1Start);
1662 if (s1Length && *s1Length <= MAX_SHORT_LINE_LENGTH) {
1663 Optional<size_t> s2Length = getLengthOnSingleLine(SM, s2Start);
1664 if (s2Length && *s2Length <= MAX_SHORT_LINE_LENGTH) {
1665 Path.erase(I);
1666 I = Path.erase(NextI);
1667 continue;
1668 }
1669 }
1670 }
1671
1672 ++I;
1673 }
1674}
1675
1676/// Return true if X is contained by Y.
1677static bool lexicalContains(const ParentMap &PM, const Stmt *X, const Stmt *Y) {
1678 while (X) {
1679 if (X == Y)
1680 return true;
1681 X = PM.getParent(X);
1682 }
1683 return false;
1684}
1685
1686// Remove short edges on the same line less than 3 columns in difference.
1687static void removePunyEdges(PathPieces &path, const SourceManager &SM,
1688 const ParentMap &PM) {
1689 bool erased = false;
1690
1691 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E;
1692 erased ? I : ++I) {
1693 erased = false;
1694
1695 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1696
1697 if (!PieceI)
1698 continue;
1699
1700 const Stmt *start = PieceI->getStartLocation().getStmtOrNull();
1701 const Stmt *end = PieceI->getEndLocation().getStmtOrNull();
1702
1703 if (!start || !end)
1704 continue;
1705
1706 const Stmt *endParent = PM.getParent(end);
1707 if (!endParent)
1708 continue;
1709
1710 if (isConditionForTerminator(end, endParent))
1711 continue;
1712
1713 SourceLocation FirstLoc = start->getBeginLoc();
1714 SourceLocation SecondLoc = end->getBeginLoc();
1715
1716 if (!SM.isWrittenInSameFile(FirstLoc, SecondLoc))
1717 continue;
1718 if (SM.isBeforeInTranslationUnit(SecondLoc, FirstLoc))
1719 std::swap(SecondLoc, FirstLoc);
1720
1721 SourceRange EdgeRange(FirstLoc, SecondLoc);
1722 Optional<size_t> ByteWidth = getLengthOnSingleLine(SM, EdgeRange);
1723
1724 // If the statements are on different lines, continue.
1725 if (!ByteWidth)
1726 continue;
1727
1728 const size_t MAX_PUNY_EDGE_LENGTH = 2;
1729 if (*ByteWidth <= MAX_PUNY_EDGE_LENGTH) {
1730 // FIXME: There are enough /bytes/ between the endpoints of the edge, but
1731 // there might not be enough /columns/. A proper user-visible column count
1732 // is probably too expensive, though.
1733 I = path.erase(I);
1734 erased = true;
1735 continue;
1736 }
1737 }
1738}
1739
1740static void removeIdenticalEvents(PathPieces &path) {
1741 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ++I) {
1742 const auto *PieceI = dyn_cast<PathDiagnosticEventPiece>(I->get());
1743
1744 if (!PieceI)
1745 continue;
1746
1747 PathPieces::iterator NextI = I; ++NextI;
1748 if (NextI == E)
1749 return;
1750
1751 const auto *PieceNextI = dyn_cast<PathDiagnosticEventPiece>(NextI->get());
1752
1753 if (!PieceNextI)
1754 continue;
1755
1756 // Erase the second piece if it has the same exact message text.
1757 if (PieceI->getString() == PieceNextI->getString()) {
1758 path.erase(NextI);
1759 }
1760 }
1761}
1762
1763static bool optimizeEdges(const PathDiagnosticConstruct &C, PathPieces &path,
1764 OptimizedCallsSet &OCS) {
1765 bool hasChanges = false;
1766 const LocationContext *LC = C.getLocationContextFor(&path);
1767 assert(LC);
1768 const ParentMap &PM = LC->getParentMap();
1769 const SourceManager &SM = C.getSourceManager();
1770
1771 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ) {
1772 // Optimize subpaths.
1773 if (auto *CallI = dyn_cast<PathDiagnosticCallPiece>(I->get())) {
1774 // Record the fact that a call has been optimized so we only do the
1775 // effort once.
1776 if (!OCS.count(CallI)) {
1777 while (optimizeEdges(C, CallI->path, OCS)) {
1778 }
1779 OCS.insert(CallI);
1780 }
1781 ++I;
1782 continue;
1783 }
1784
1785 // Pattern match the current piece and its successor.
1786 auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get());
1787
1788 if (!PieceI) {
1789 ++I;
1790 continue;
1791 }
1792
1793 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull();
1794 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull();
1795 const Stmt *level1 = getStmtParent(s1Start, PM);
1796 const Stmt *level2 = getStmtParent(s1End, PM);
1797
1798 PathPieces::iterator NextI = I; ++NextI;
1799 if (NextI == E)
1800 break;
1801
1802 const auto *PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get());
1803
1804 if (!PieceNextI) {
1805 ++I;
1806 continue;
1807 }
1808
1809 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull();
1810 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull();
1811 const Stmt *level3 = getStmtParent(s2Start, PM);
1812 const Stmt *level4 = getStmtParent(s2End, PM);
1813
1814 // Rule I.
1815 //
1816 // If we have two consecutive control edges whose end/begin locations
1817 // are at the same level (e.g. statements or top-level expressions within
1818 // a compound statement, or siblings share a single ancestor expression),
1819 // then merge them if they have no interesting intermediate event.
1820 //
1821 // For example:
1822 //
1823 // (1.1 -> 1.2) -> (1.2 -> 1.3) becomes (1.1 -> 1.3) because the common
1824 // parent is '1'. Here 'x.y.z' represents the hierarchy of statements.
1825 //
1826 // NOTE: this will be limited later in cases where we add barriers
1827 // to prevent this optimization.
1828 if (level1 && level1 == level2 && level1 == level3 && level1 == level4) {
1829 PieceI->setEndLocation(PieceNextI->getEndLocation());
1830 path.erase(NextI);
1831 hasChanges = true;
1832 continue;
1833 }
1834
1835 // Rule II.
1836 //
1837 // Eliminate edges between subexpressions and parent expressions
1838 // when the subexpression is consumed.
1839 //
1840 // NOTE: this will be limited later in cases where we add barriers
1841 // to prevent this optimization.
1842 if (s1End && s1End == s2Start && level2) {
1843 bool removeEdge = false;
1844 // Remove edges into the increment or initialization of a
1845 // loop that have no interleaving event. This means that
1846 // they aren't interesting.
1847 if (isIncrementOrInitInForLoop(s1End, level2))
1848 removeEdge = true;
1849 // Next only consider edges that are not anchored on
1850 // the condition of a terminator. This are intermediate edges
1851 // that we might want to trim.
1852 else if (!isConditionForTerminator(level2, s1End)) {
1853 // Trim edges on expressions that are consumed by
1854 // the parent expression.
1855 if (isa<Expr>(s1End) && PM.isConsumedExpr(cast<Expr>(s1End))) {
1856 removeEdge = true;
1857 }
1858 // Trim edges where a lexical containment doesn't exist.
1859 // For example:
1860 //
1861 // X -> Y -> Z
1862 //
1863 // If 'Z' lexically contains Y (it is an ancestor) and
1864 // 'X' does not lexically contain Y (it is a descendant OR
1865 // it has no lexical relationship at all) then trim.
1866 //
1867 // This can eliminate edges where we dive into a subexpression
1868 // and then pop back out, etc.
1869 else if (s1Start && s2End &&
1870 lexicalContains(PM, s2Start, s2End) &&
1871 !lexicalContains(PM, s1End, s1Start)) {
1872 removeEdge = true;
1873 }
1874 // Trim edges from a subexpression back to the top level if the
1875 // subexpression is on a different line.
1876 //
1877 // A.1 -> A -> B
1878 // becomes
1879 // A.1 -> B
1880 //
1881 // These edges just look ugly and don't usually add anything.
1882 else if (s1Start && s2End &&
1883 lexicalContains(PM, s1Start, s1End)) {
1884 SourceRange EdgeRange(PieceI->getEndLocation().asLocation(),
1885 PieceI->getStartLocation().asLocation());
1886 if (!getLengthOnSingleLine(SM, EdgeRange))
1887 removeEdge = true;
1888 }
1889 }
1890
1891 if (removeEdge) {
1892 PieceI->setEndLocation(PieceNextI->getEndLocation());
1893 path.erase(NextI);
1894 hasChanges = true;
1895 continue;
1896 }
1897 }
1898
1899 // Optimize edges for ObjC fast-enumeration loops.
1900 //
1901 // (X -> collection) -> (collection -> element)
1902 //
1903 // becomes:
1904 //
1905 // (X -> element)
1906 if (s1End == s2Start) {
1907 const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(level3);
1908 if (FS && FS->getCollection()->IgnoreParens() == s2Start &&
1909 s2End == FS->getElement()) {
1910 PieceI->setEndLocation(PieceNextI->getEndLocation());
1911 path.erase(NextI);
1912 hasChanges = true;
1913 continue;
1914 }
1915 }
1916
1917 // No changes at this index? Move to the next one.
1918 ++I;
1919 }
1920
1921 if (!hasChanges) {
1922 // Adjust edges into subexpressions to make them more uniform
1923 // and aesthetically pleasing.
1924 addContextEdges(path, LC);
1925 // Remove "cyclical" edges that include one or more context edges.
1926 removeContextCycles(path, SM);
1927 // Hoist edges originating from branch conditions to branches
1928 // for simple branches.
1929 simplifySimpleBranches(path);
1930 // Remove any puny edges left over after primary optimization pass.
1931 removePunyEdges(path, SM, PM);
1932 // Remove identical events.
1933 removeIdenticalEvents(path);
1934 }
1935
1936 return hasChanges;
1937}
1938
1939/// Drop the very first edge in a path, which should be a function entry edge.
1940///
1941/// If the first edge is not a function entry edge (say, because the first
1942/// statement had an invalid source location), this function does nothing.
1943// FIXME: We should just generate invalid edges anyway and have the optimizer
1944// deal with them.
1945static void dropFunctionEntryEdge(const PathDiagnosticConstruct &C,
1946 PathPieces &Path) {
1947 const auto *FirstEdge =
1948 dyn_cast<PathDiagnosticControlFlowPiece>(Path.front().get());
1949 if (!FirstEdge)
1950 return;
1951
1952 const Decl *D = C.getLocationContextFor(&Path)->getDecl();
1953 PathDiagnosticLocation EntryLoc =
1954 PathDiagnosticLocation::createBegin(D, C.getSourceManager());
1955 if (FirstEdge->getStartLocation() != EntryLoc)
1956 return;
1957
1958 Path.pop_front();
1959}
1960
1961/// Populate executes lines with lines containing at least one diagnostics.
1962static void updateExecutedLinesWithDiagnosticPieces(PathDiagnostic &PD) {
1963
1964 PathPieces path = PD.path.flatten(/*ShouldFlattenMacros=*/true);
1965 FilesToLineNumsMap &ExecutedLines = PD.getExecutedLines();
1966
1967 for (const auto &P : path) {
1968 FullSourceLoc Loc = P->getLocation().asLocation().getExpansionLoc();
1969 FileID FID = Loc.getFileID();
1970 unsigned LineNo = Loc.getLineNumber();
1971 assert(FID.isValid());
1972 ExecutedLines[FID].insert(LineNo);
1973 }
1974}
1975
1976PathDiagnosticConstruct::PathDiagnosticConstruct(
1977 const PathDiagnosticConsumer *PDC, const ExplodedNode *ErrorNode,
1978 const PathSensitiveBugReport *R)
1979 : Consumer(PDC), CurrentNode(ErrorNode),
1980 SM(CurrentNode->getCodeDecl().getASTContext().getSourceManager()),
1981 PD(generateEmptyDiagnosticForReport(R, getSourceManager())) {
1982 LCM[&PD->getActivePath()] = ErrorNode->getLocationContext();
1983}
1984
1985PathDiagnosticBuilder::PathDiagnosticBuilder(
1986 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath,
1987 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode,
1988 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics)
1989 : BugReporterContext(BRC), BugPath(std::move(BugPath)), R(r),
1990 ErrorNode(ErrorNode),
1991 VisitorsDiagnostics(std::move(VisitorsDiagnostics)) {}
1992
1993std::unique_ptr<PathDiagnostic>
1994PathDiagnosticBuilder::generate(const PathDiagnosticConsumer *PDC) const {
1995 PathDiagnosticConstruct Construct(PDC, ErrorNode, R);
1996
1997 const SourceManager &SM = getSourceManager();
1998 const AnalyzerOptions &Opts = getAnalyzerOptions();
1999
2000 if (!PDC->shouldGenerateDiagnostics())
2001 return generateEmptyDiagnosticForReport(R, getSourceManager());
2002
2003 // Construct the final (warning) event for the bug report.
2004 auto EndNotes = VisitorsDiagnostics->find(ErrorNode);
2005 PathDiagnosticPieceRef LastPiece;
2006 if (EndNotes != VisitorsDiagnostics->end()) {
2007 assert(!EndNotes->second.empty());
2008 LastPiece = EndNotes->second[0];
2009 } else {
2010 LastPiece = BugReporterVisitor::getDefaultEndPath(*this, ErrorNode,
2011 *getBugReport());
2012 }
2013 Construct.PD->setEndOfPath(LastPiece);
2014
2015 PathDiagnosticLocation PrevLoc = Construct.PD->getLocation();
2016 // From the error node to the root, ascend the bug path and construct the bug
2017 // report.
2018 while (Construct.ascendToPrevNode()) {
2019 generatePathDiagnosticsForNode(Construct, PrevLoc);
2020
2021 auto VisitorNotes = VisitorsDiagnostics->find(Construct.getCurrentNode());
2022 if (VisitorNotes == VisitorsDiagnostics->end())
2023 continue;
2024
2025 // This is a workaround due to inability to put shared PathDiagnosticPiece
2026 // into a FoldingSet.
2027 std::set<llvm::FoldingSetNodeID> DeduplicationSet;
2028
2029 // Add pieces from custom visitors.
2030 for (const PathDiagnosticPieceRef &Note : VisitorNotes->second) {
2031 llvm::FoldingSetNodeID ID;
2032 Note->Profile(ID);
2033 if (!DeduplicationSet.insert(ID).second)
2034 continue;
2035
2036 if (PDC->shouldAddPathEdges())
2037 addEdgeToPath(Construct.getActivePath(), PrevLoc, Note->getLocation());
2038 updateStackPiecesWithMessage(Note, Construct.CallStack);
2039 Construct.getActivePath().push_front(Note);
2040 }
2041 }
2042
2043 if (PDC->shouldAddPathEdges()) {
2044 // Add an edge to the start of the function.
2045 // We'll prune it out later, but it helps make diagnostics more uniform.
2046 const StackFrameContext *CalleeLC =
2047 Construct.getLocationContextForActivePath()->getStackFrame();
2048 const Decl *D = CalleeLC->getDecl();
2049 addEdgeToPath(Construct.getActivePath(), PrevLoc,
2050 PathDiagnosticLocation::createBegin(D, SM));
2051 }
2052
2053
2054 // Finally, prune the diagnostic path of uninteresting stuff.
2055 if (!Construct.PD->path.empty()) {
2056 if (R->shouldPrunePath() && Opts.ShouldPrunePaths) {
2057 bool stillHasNotes =
2058 removeUnneededCalls(Construct, Construct.getMutablePieces(), R);
2059 assert(stillHasNotes);
2060 (void)stillHasNotes;
2061 }
2062
2063 // Remove pop-up notes if needed.
2064 if (!Opts.ShouldAddPopUpNotes)
2065 removePopUpNotes(Construct.getMutablePieces());
2066
2067 // Redirect all call pieces to have valid locations.
2068 adjustCallLocations(Construct.getMutablePieces());
2069 removePiecesWithInvalidLocations(Construct.getMutablePieces());
2070
2071 if (PDC->shouldAddPathEdges()) {
2072
2073 // Reduce the number of edges from a very conservative set
2074 // to an aesthetically pleasing subset that conveys the
2075 // necessary information.
2076 OptimizedCallsSet OCS;
2077 while (optimizeEdges(Construct, Construct.getMutablePieces(), OCS)) {
2078 }
2079
2080 // Drop the very first function-entry edge. It's not really necessary
2081 // for top-level functions.
2082 dropFunctionEntryEdge(Construct, Construct.getMutablePieces());
2083 }
2084
2085 // Remove messages that are basically the same, and edges that may not
2086 // make sense.
2087 // We have to do this after edge optimization in the Extensive mode.
2088 removeRedundantMsgs(Construct.getMutablePieces());
2089 removeEdgesToDefaultInitializers(Construct.getMutablePieces());
2090 }
2091
2092 if (Opts.ShouldDisplayMacroExpansions)
2093 CompactMacroExpandedPieces(Construct.getMutablePieces(), SM);
2094
2095 return std::move(Construct.PD);
2096}
2097
2098//===----------------------------------------------------------------------===//
2099// Methods for BugType and subclasses.
2100//===----------------------------------------------------------------------===//
2101
2102void BugType::anchor() {}
2103
2104void BuiltinBug::anchor() {}
2105
2106//===----------------------------------------------------------------------===//
2107// Methods for BugReport and subclasses.
2108//===----------------------------------------------------------------------===//
2109
2110LLVM_ATTRIBUTE_USED static bool
2111isDependency(const CheckerRegistryData &Registry, StringRef CheckerName) {
2112 for (const std::pair<StringRef, StringRef> &Pair : Registry.Dependencies) {
2113 if (Pair.second == CheckerName)
2114 return true;
2115 }
2116 return false;
2117}
2118
2119LLVM_ATTRIBUTE_USED static bool isHidden(const CheckerRegistryData &Registry,
2120 StringRef CheckerName) {
2121 for (const CheckerInfo &Checker : Registry.Checkers) {
2122 if (Checker.FullName == CheckerName)
2123 return Checker.IsHidden;
2124 }
2125 llvm_unreachable(
2126 "Checker name not found in CheckerRegistry -- did you retrieve it "
2127 "correctly from CheckerManager::getCurrentCheckerName?");
2128}
2129
2130PathSensitiveBugReport::PathSensitiveBugReport(
2131 const BugType &bt, StringRef shortDesc, StringRef desc,
2132 const ExplodedNode *errorNode, PathDiagnosticLocation LocationToUnique,
2133 const Decl *DeclToUnique)
2134 : BugReport(Kind::PathSensitive, bt, shortDesc, desc), ErrorNode(errorNode),
2135 ErrorNodeRange(getStmt() ? getStmt()->getSourceRange() : SourceRange()),
2136 UniqueingLocation(LocationToUnique), UniqueingDecl(DeclToUnique) {
2137 assert(!isDependency(ErrorNode->getState()
2138 ->getAnalysisManager()
2139 .getCheckerManager()
2140 ->getCheckerRegistryData(),
2141 bt.getCheckerName()) &&
2142 "Some checkers depend on this one! We don't allow dependency "
2143 "checkers to emit warnings, because checkers should depend on "
2144 "*modeling*, not *diagnostics*.");
2145
2146 assert(
2147 (bt.getCheckerName().startswith("debug") ||
2148 !isHidden(ErrorNode->getState()
2149 ->getAnalysisManager()
2150 .getCheckerManager()
2151 ->getCheckerRegistryData(),
2152 bt.getCheckerName())) &&
2153 "Hidden checkers musn't emit diagnostics as they are by definition "
2154 "non-user facing!");
2155}
2156
2157void PathSensitiveBugReport::addVisitor(
2158 std::unique_ptr<BugReporterVisitor> visitor) {
2159 if (!visitor)
2160 return;
2161
2162 llvm::FoldingSetNodeID ID;
2163 visitor->Profile(ID);
2164
2165 void *InsertPos = nullptr;
2166 if (CallbacksSet.FindNodeOrInsertPos(ID, InsertPos)) {
2167 return;
2168 }
2169
2170 Callbacks.push_back(std::move(visitor));
2171}
2172
2173void PathSensitiveBugReport::clearVisitors() {
2174 Callbacks.clear();
2175}
2176
2177const Decl *PathSensitiveBugReport::getDeclWithIssue() const {
2178 const ExplodedNode *N = getErrorNode();
2179 if (!N)
2180 return nullptr;
2181
2182 const LocationContext *LC = N->getLocationContext();
2183 return LC->getStackFrame()->getDecl();
2184}
2185
2186void BasicBugReport::Profile(llvm::FoldingSetNodeID& hash) const {
2187 hash.AddInteger(static_cast<int>(getKind()));
2188 hash.AddPointer(&BT);
2189 hash.AddString(Description);
2190 assert(Location.isValid());
2191 Location.Profile(hash);
2192
2193 for (SourceRange range : Ranges) {
2194 if (!range.isValid())
2195 continue;
2196 hash.Add(range.getBegin());
2197 hash.Add(range.getEnd());
2198 }
2199}
2200
2201void PathSensitiveBugReport::Profile(llvm::FoldingSetNodeID &hash) const {
2202 hash.AddInteger(static_cast<int>(getKind()));
2203 hash.AddPointer(&BT);
2204 hash.AddString(Description);
2205 PathDiagnosticLocation UL = getUniqueingLocation();
2206 if (UL.isValid()) {
2207 UL.Profile(hash);
2208 } else {
2209 // TODO: The statement may be null if the report was emitted before any
2210 // statements were executed. In particular, some checkers by design
2211 // occasionally emit their reports in empty functions (that have no
2212 // statements in their body). Do we profile correctly in this case?
2213 hash.AddPointer(ErrorNode->getCurrentOrPreviousStmtForDiagnostics());
2214 }
2215
2216 for (SourceRange range : Ranges) {
2217 if (!range.isValid())
2218 continue;
2219 hash.Add(range.getBegin());
2220 hash.Add(range.getEnd());
2221 }
2222}
2223
2224template <class T>
2225static void insertToInterestingnessMap(
2226 llvm::DenseMap<T, bugreporter::TrackingKind> &InterestingnessMap, T Val,
2227 bugreporter::TrackingKind TKind) {
2228 auto Result = InterestingnessMap.insert({Val, TKind});
2229
2230 if (Result.second)
2231 return;
2232
2233 // Even if this symbol/region was already marked as interesting as a
2234 // condition, if we later mark it as interesting again but with
2235 // thorough tracking, overwrite it. Entities marked with thorough
2236 // interestiness are the most important (or most interesting, if you will),
2237 // and we wouldn't like to downplay their importance.
2238
2239 switch (TKind) {
2240 case bugreporter::TrackingKind::Thorough:
2241 Result.first->getSecond() = bugreporter::TrackingKind::Thorough;
2242 return;
2243 case bugreporter::TrackingKind::Condition:
2244 return;
2245 }
2246
2247 llvm_unreachable(
2248 "BugReport::markInteresting currently can only handle 2 different "
2249 "tracking kinds! Please define what tracking kind should this entitiy"
2250 "have, if it was already marked as interesting with a different kind!");
2251}
2252
2253void PathSensitiveBugReport::markInteresting(SymbolRef sym,
2254 bugreporter::TrackingKind TKind) {
2255 if (!sym)
2256 return;
2257
2258 insertToInterestingnessMap(InterestingSymbols, sym, TKind);
2259
2260 // FIXME: No tests exist for this code and it is questionable:
2261 // How to handle multiple metadata for the same region?
2262 if (const auto *meta = dyn_cast<SymbolMetadata>(sym))
2263 markInteresting(meta->getRegion(), TKind);
2264}
2265
2266void PathSensitiveBugReport::markNotInteresting(SymbolRef sym) {
2267 if (!sym)
2268 return;
2269 InterestingSymbols.erase(sym);
2270
2271 // The metadata part of markInteresting is not reversed here.
2272 // Just making the same region not interesting is incorrect
2273 // in specific cases.
2274 if (const auto *meta = dyn_cast<SymbolMetadata>(sym))
2275 markNotInteresting(meta->getRegion());
2276}
2277
2278void PathSensitiveBugReport::markInteresting(const MemRegion *R,
2279 bugreporter::TrackingKind TKind) {
2280 if (!R)
2281 return;
2282
2283 R = R->getBaseRegion();
2284 insertToInterestingnessMap(InterestingRegions, R, TKind);
2285
2286 if (const auto *SR = dyn_cast<SymbolicRegion>(R))
2287 markInteresting(SR->getSymbol(), TKind);
2288}
2289
2290void PathSensitiveBugReport::markNotInteresting(const MemRegion *R) {
2291 if (!R)
2292 return;
2293
2294 R = R->getBaseRegion();
2295 InterestingRegions.erase(R);
2296
2297 if (const auto *SR = dyn_cast<SymbolicRegion>(R))
2298 markNotInteresting(SR->getSymbol());
2299}
2300
2301void PathSensitiveBugReport::markInteresting(SVal V,
2302 bugreporter::TrackingKind TKind) {
2303 markInteresting(V.getAsRegion(), TKind);
2304 markInteresting(V.getAsSymbol(), TKind);
2305}
2306
2307void PathSensitiveBugReport::markInteresting(const LocationContext *LC) {
2308 if (!LC)
2309 return;
2310 InterestingLocationContexts.insert(LC);
2311}
2312
2313Optional<bugreporter::TrackingKind>
2314PathSensitiveBugReport::getInterestingnessKind(SVal V) const {
2315 auto RKind = getInterestingnessKind(V.getAsRegion());
2316 auto SKind = getInterestingnessKind(V.getAsSymbol());
2317 if (!RKind)
2318 return SKind;
2319 if (!SKind)
2320 return RKind;
2321
2322 // If either is marked with throrough tracking, return that, we wouldn't like
2323 // to downplay a note's importance by 'only' mentioning it as a condition.
2324 switch(*RKind) {
2325 case bugreporter::TrackingKind::Thorough:
2326 return RKind;
2327 case bugreporter::TrackingKind::Condition:
2328 return SKind;
2329 }
2330
2331 llvm_unreachable(
2332 "BugReport::getInterestingnessKind currently can only handle 2 different "
2333 "tracking kinds! Please define what tracking kind should we return here "
2334 "when the kind of getAsRegion() and getAsSymbol() is different!");
2335 return None;
2336}
2337
2338Optional<bugreporter::TrackingKind>
2339PathSensitiveBugReport::getInterestingnessKind(SymbolRef sym) const {
2340 if (!sym)
2341 return None;
2342 // We don't currently consider metadata symbols to be interesting
2343 // even if we know their region is interesting. Is that correct behavior?
2344 auto It = InterestingSymbols.find(sym);
2345 if (It == InterestingSymbols.end())
2346 return None;
2347 return It->getSecond();
2348}
2349
2350Optional<bugreporter::TrackingKind>
2351PathSensitiveBugReport::getInterestingnessKind(const MemRegion *R) const {
2352 if (!R)
2353 return None;
2354
2355 R = R->getBaseRegion();
2356 auto It = InterestingRegions.find(R);
2357 if (It != InterestingRegions.end())
2358 return It->getSecond();
2359
2360 if (const auto *SR = dyn_cast<SymbolicRegion>(R))
2361 return getInterestingnessKind(SR->getSymbol());
2362 return None;
2363}
2364
2365bool PathSensitiveBugReport::isInteresting(SVal V) const {
2366 return getInterestingnessKind(V).has_value();
2367}
2368
2369bool PathSensitiveBugReport::isInteresting(SymbolRef sym) const {
2370 return getInterestingnessKind(sym).has_value();
2371}
2372
2373bool PathSensitiveBugReport::isInteresting(const MemRegion *R) const {
2374 return getInterestingnessKind(R).has_value();
2375}
2376
2377bool PathSensitiveBugReport::isInteresting(const LocationContext *LC) const {
2378 if (!LC)
2379 return false;
2380 return InterestingLocationContexts.count(LC);
2381}
2382
2383const Stmt *PathSensitiveBugReport::getStmt() const {
2384 if (!ErrorNode)
2385 return nullptr;
2386
2387 ProgramPoint ProgP = ErrorNode->getLocation();
2388 const Stmt *S = nullptr;
2389
2390 if (Optional<BlockEntrance> BE = ProgP.getAs<BlockEntrance>()) {
2391 CFGBlock &Exit = ProgP.getLocationContext()->getCFG()->getExit();
2392 if (BE->getBlock() == &Exit)
2393 S = ErrorNode->getPreviousStmtForDiagnostics();
2394 }
2395 if (!S)
2396 S = ErrorNode->getStmtForDiagnostics();
2397
2398 return S;
2399}
2400
2401ArrayRef<SourceRange>
2402PathSensitiveBugReport::getRanges() const {
2403 // If no custom ranges, add the range of the statement corresponding to
2404 // the error node.
2405 if (Ranges.empty() && isa_and_nonnull<Expr>(getStmt()))
2406 return ErrorNodeRange;
2407
2408 return Ranges;
2409}
2410
2411PathDiagnosticLocation
2412PathSensitiveBugReport::getLocation() const {
2413 assert(ErrorNode && "Cannot create a location with a null node.");
2414 const Stmt *S = ErrorNode->getStmtForDiagnostics();
2415 ProgramPoint P = ErrorNode->getLocation();
2416 const LocationContext *LC = P.getLocationContext();
2417 SourceManager &SM =
2418 ErrorNode->getState()->getStateManager().getContext().getSourceManager();
2419
2420 if (!S) {
2421 // If this is an implicit call, return the implicit call point location.
2422 if (Optional<PreImplicitCall> PIE = P.getAs<PreImplicitCall>())
2423 return PathDiagnosticLocation(PIE->getLocation(), SM);
2424 if (auto FE = P.getAs<FunctionExitPoint>()) {
2425 if (const ReturnStmt *RS = FE->getStmt())
2426 return PathDiagnosticLocation::createBegin(RS, SM, LC);
2427 }
2428 S = ErrorNode->getNextStmtForDiagnostics();
2429 }
2430
2431 if (S) {
2432 // For member expressions, return the location of the '.' or '->'.
2433 if (const auto *ME = dyn_cast<MemberExpr>(S))
2434 return PathDiagnosticLocation::createMemberLoc(ME, SM);
2435
2436 // For binary operators, return the location of the operator.
2437 if (const auto *B = dyn_cast<BinaryOperator>(S))
2438 return PathDiagnosticLocation::createOperatorLoc(B, SM);
2439
2440 if (P.getAs<PostStmtPurgeDeadSymbols>())
2441 return PathDiagnosticLocation::createEnd(S, SM, LC);
2442
2443 if (S->getBeginLoc().isValid())
2444 return PathDiagnosticLocation(S, SM, LC);
2445
2446 return PathDiagnosticLocation(
2447 PathDiagnosticLocation::getValidSourceLocation(S, LC), SM);
2448 }
2449
2450 return PathDiagnosticLocation::createDeclEnd(ErrorNode->getLocationContext(),
2451 SM);
2452}
2453
2454//===----------------------------------------------------------------------===//
2455// Methods for BugReporter and subclasses.
2456//===----------------------------------------------------------------------===//
2457
2458const ExplodedGraph &PathSensitiveBugReporter::getGraph() const {
2459 return Eng.getGraph();
2460}
2461
2462ProgramStateManager &PathSensitiveBugReporter::getStateManager() const {
2463 return Eng.getStateManager();
2464}
2465
2466BugReporter::BugReporter(BugReporterData &d) : D(d) {}
2467BugReporter::~BugReporter() {
2468 // Make sure reports are flushed.
2469 assert(StrBugTypes.empty() &&
2470 "Destroying BugReporter before diagnostics are emitted!");
2471
2472 // Free the bug reports we are tracking.
2473 for (const auto I : EQClassesVector)
2474 delete I;
2475}
2476
2477void BugReporter::FlushReports() {
2478 // We need to flush reports in deterministic order to ensure the order
2479 // of the reports is consistent between runs.
2480 for (const auto EQ : EQClassesVector)
2481 FlushReport(*EQ);
2482
2483 // BugReporter owns and deletes only BugTypes created implicitly through
2484 // EmitBasicReport.
2485 // FIXME: There are leaks from checkers that assume that the BugTypes they
2486 // create will be destroyed by the BugReporter.
2487 StrBugTypes.clear();
2488}
2489
2490//===----------------------------------------------------------------------===//
2491// PathDiagnostics generation.
2492//===----------------------------------------------------------------------===//
2493
2494namespace {
2495
2496/// A wrapper around an ExplodedGraph that contains a single path from the root
2497/// to the error node.
2498class BugPathInfo {
2499public:
2500 std::unique_ptr<ExplodedGraph> BugPath;
2501 PathSensitiveBugReport *Report;
2502 const ExplodedNode *ErrorNode;
2503};
2504
2505/// A wrapper around an ExplodedGraph whose leafs are all error nodes. Can
2506/// conveniently retrieve bug paths from a single error node to the root.
2507class BugPathGetter {
2508 std::unique_ptr<ExplodedGraph> TrimmedGraph;
2509
2510 using PriorityMapTy = llvm::DenseMap<const ExplodedNode *, unsigned>;
2511
2512 /// Assign each node with its distance from the root.
2513 PriorityMapTy PriorityMap;
2514
2515 /// Since the getErrorNode() or BugReport refers to the original ExplodedGraph,
2516 /// we need to pair it to the error node of the constructed trimmed graph.
2517 using ReportNewNodePair =
2518 std::pair<PathSensitiveBugReport *, const ExplodedNode *>;
2519 SmallVector<ReportNewNodePair, 32> ReportNodes;
2520
2521 BugPathInfo CurrentBugPath;
2522
2523 /// A helper class for sorting ExplodedNodes by priority.
2524 template <bool Descending>
2525 class PriorityCompare {
2526 const PriorityMapTy &PriorityMap;
2527
2528 public:
2529 PriorityCompare(const PriorityMapTy &M) : PriorityMap(M) {}
2530
2531 bool operator()(const ExplodedNode *LHS, const ExplodedNode *RHS) const {
2532 PriorityMapTy::const_iterator LI = PriorityMap.find(LHS);
2533 PriorityMapTy::const_iterator RI = PriorityMap.find(RHS);
2534 PriorityMapTy::const_iterator E = PriorityMap.end();
2535
2536 if (LI == E)
2537 return Descending;
2538 if (RI == E)
2539 return !Descending;
2540
2541 return Descending ? LI->second > RI->second
2542 : LI->second < RI->second;
2543 }
2544
2545 bool operator()(const ReportNewNodePair &LHS,
2546 const ReportNewNodePair &RHS) const {
2547 return (*this)(LHS.second, RHS.second);
2548 }
2549 };
2550
2551public:
2552 BugPathGetter(const ExplodedGraph *OriginalGraph,
2553 ArrayRef<PathSensitiveBugReport *> &bugReports);
2554
2555 BugPathInfo *getNextBugPath();
2556};
2557
2558} // namespace
2559
2560BugPathGetter::BugPathGetter(const ExplodedGraph *OriginalGraph,
2561 ArrayRef<PathSensitiveBugReport *> &bugReports) {
2562 SmallVector<const ExplodedNode *, 32> Nodes;
2563 for (const auto I : bugReports) {
2564 assert(I->isValid() &&
2565 "We only allow BugReporterVisitors and BugReporter itself to "
2566 "invalidate reports!");
2567 Nodes.emplace_back(I->getErrorNode());
2568 }
2569
2570 // The trimmed graph is created in the body of the constructor to ensure
2571 // that the DenseMaps have been initialized already.
2572 InterExplodedGraphMap ForwardMap;
2573 TrimmedGraph = OriginalGraph->trim(Nodes, &ForwardMap);
2574
2575 // Find the (first) error node in the trimmed graph. We just need to consult
2576 // the node map which maps from nodes in the original graph to nodes
2577 // in the new graph.
2578 llvm::SmallPtrSet<const ExplodedNode *, 32> RemainingNodes;
2579
2580 for (PathSensitiveBugReport *Report : bugReports) {
2581 const ExplodedNode *NewNode = ForwardMap.lookup(Report->getErrorNode());
2582 assert(NewNode &&
2583 "Failed to construct a trimmed graph that contains this error "
2584 "node!");
2585 ReportNodes.emplace_back(Report, NewNode);
2586 RemainingNodes.insert(NewNode);
2587 }
2588
2589 assert(!RemainingNodes.empty() && "No error node found in the trimmed graph");
2590
2591 // Perform a forward BFS to find all the shortest paths.
2592 std::queue<const ExplodedNode *> WS;
2593
2594 assert(TrimmedGraph->num_roots() == 1);
2595 WS.push(*TrimmedGraph->roots_begin());
2596 unsigned Priority = 0;
2597
2598 while (!WS.empty()) {
2599 const ExplodedNode *Node = WS.front();
2600 WS.pop();
2601
2602 PriorityMapTy::iterator PriorityEntry;
2603 bool IsNew;
2604 std::tie(PriorityEntry, IsNew) = PriorityMap.insert({Node, Priority});
2605 ++Priority;
2606
2607 if (!IsNew) {
2608 assert(PriorityEntry->second <= Priority);
2609 continue;
2610 }
2611
2612 if (RemainingNodes.erase(Node))
2613 if (RemainingNodes.empty())
2614 break;
2615
2616 for (const ExplodedNode *Succ : Node->succs())
2617 WS.push(Succ);
2618 }
2619
2620 // Sort the error paths from longest to shortest.
2621 llvm::sort(ReportNodes, PriorityCompare<true>(PriorityMap));
2622}
2623
2624BugPathInfo *BugPathGetter::getNextBugPath() {
2625 if (ReportNodes.empty())
2626 return nullptr;
2627
2628 const ExplodedNode *OrigN;
2629 std::tie(CurrentBugPath.Report, OrigN) = ReportNodes.pop_back_val();
2630 assert(PriorityMap.find(OrigN) != PriorityMap.end() &&
2631 "error node not accessible from root");
2632
2633 // Create a new graph with a single path. This is the graph that will be
2634 // returned to the caller.
2635 auto GNew = std::make_unique<ExplodedGraph>();
2636
2637 // Now walk from the error node up the BFS path, always taking the
2638 // predeccessor with the lowest number.
2639 ExplodedNode *Succ = nullptr;
2640 while (true) {
2641 // Create the equivalent node in the new graph with the same state
2642 // and location.
2643 ExplodedNode *NewN = GNew->createUncachedNode(
2644 OrigN->getLocation(), OrigN->getState(),
2645 OrigN->getID(), OrigN->isSink());
2646
2647 // Link up the new node with the previous node.
2648 if (Succ)
2649 Succ->addPredecessor(NewN, *GNew);
2650 else
2651 CurrentBugPath.ErrorNode = NewN;
2652
2653 Succ = NewN;
2654
2655 // Are we at the final node?
2656 if (OrigN->pred_empty()) {
2657 GNew->addRoot(NewN);
2658 break;
2659 }
2660
2661 // Find the next predeccessor node. We choose the node that is marked
2662 // with the lowest BFS number.
2663 OrigN = *std::min_element(OrigN->pred_begin(), OrigN->pred_end(),
2664 PriorityCompare<false>(PriorityMap));
2665 }
2666
2667 CurrentBugPath.BugPath = std::move(GNew);
2668
2669 return &CurrentBugPath;
2670}
2671
2672/// CompactMacroExpandedPieces - This function postprocesses a PathDiagnostic
2673/// object and collapses PathDiagosticPieces that are expanded by macros.
2674static void CompactMacroExpandedPieces(PathPieces &path,
2675 const SourceManager& SM) {
2676 using MacroStackTy = std::vector<
2677 std::pair<std::shared_ptr<PathDiagnosticMacroPiece>, SourceLocation>>;
2678
2679 using PiecesTy = std::vector<PathDiagnosticPieceRef>;
2680
2681 MacroStackTy MacroStack;
2682 PiecesTy Pieces;
2683
2684 for (PathPieces::const_iterator I = path.begin(), E = path.end();
2685 I != E; ++I) {
2686 const auto &piece = *I;
2687
2688 // Recursively compact calls.
2689 if (auto *call = dyn_cast<PathDiagnosticCallPiece>(&*piece)) {
2690 CompactMacroExpandedPieces(call->path, SM);
2691 }
2692
2693 // Get the location of the PathDiagnosticPiece.
2694 const FullSourceLoc Loc = piece->getLocation().asLocation();
2695
2696 // Determine the instantiation location, which is the location we group
2697 // related PathDiagnosticPieces.
2698 SourceLocation InstantiationLoc = Loc.isMacroID() ?
2699 SM.getExpansionLoc(Loc) :
2700 SourceLocation();
2701
2702 if (Loc.isFileID()) {
2703 MacroStack.clear();
2704 Pieces.push_back(piece);
2705 continue;
2706 }
2707
2708 assert(Loc.isMacroID());
2709
2710 // Is the PathDiagnosticPiece within the same macro group?
2711 if (!MacroStack.empty() && InstantiationLoc == MacroStack.back().second) {
2712 MacroStack.back().first->subPieces.push_back(piece);
2713 continue;
2714 }
2715
2716 // We aren't in the same group. Are we descending into a new macro
2717 // or are part of an old one?
2718 std::shared_ptr<PathDiagnosticMacroPiece> MacroGroup;
2719
2720 SourceLocation ParentInstantiationLoc = InstantiationLoc.isMacroID() ?
2721 SM.getExpansionLoc(Loc) :
2722 SourceLocation();
2723
2724 // Walk the entire macro stack.
2725 while (!MacroStack.empty()) {
2726 if (InstantiationLoc == MacroStack.back().second) {
2727 MacroGroup = MacroStack.back().first;
2728 break;
2729 }
2730
2731 if (ParentInstantiationLoc == MacroStack.back().second) {
2732 MacroGroup = MacroStack.back().first;
2733 break;
2734 }
2735
2736 MacroStack.pop_back();
2737 }
2738
2739 if (!MacroGroup || ParentInstantiationLoc == MacroStack.back().second) {
2740 // Create a new macro group and add it to the stack.
2741 auto NewGroup = std::make_shared<PathDiagnosticMacroPiece>(
2742 PathDiagnosticLocation::createSingleLocation(piece->getLocation()));
2743
2744 if (MacroGroup)
2745 MacroGroup->subPieces.push_back(NewGroup);
2746 else {
2747 assert(InstantiationLoc.isFileID());
2748 Pieces.push_back(NewGroup);
2749 }
2750
2751 MacroGroup = NewGroup;
2752 MacroStack.push_back(std::make_pair(MacroGroup, InstantiationLoc));
2753 }
2754
2755 // Finally, add the PathDiagnosticPiece to the group.
2756 MacroGroup->subPieces.push_back(piece);
2757 }
2758
2759 // Now take the pieces and construct a new PathDiagnostic.
2760 path.clear();
2761
2762 path.insert(path.end(), Pieces.begin(), Pieces.end());
2763}
2764
2765/// Generate notes from all visitors.
2766/// Notes associated with @c ErrorNode are generated using
2767/// @c getEndPath, and the rest are generated with @c VisitNode.
2768static std::unique_ptr<VisitorsDiagnosticsTy>
2769generateVisitorsDiagnostics(PathSensitiveBugReport *R,
2770 const ExplodedNode *ErrorNode,
2771 BugReporterContext &BRC) {
2772 std::unique_ptr<VisitorsDiagnosticsTy> Notes =
2773 std::make_unique<VisitorsDiagnosticsTy>();
2774 PathSensitiveBugReport::VisitorList visitors;
2775
2776 // Run visitors on all nodes starting from the node *before* the last one.
2777 // The last node is reserved for notes generated with @c getEndPath.
2778 const ExplodedNode *NextNode = ErrorNode->getFirstPred();
2779 while (NextNode) {
2780
2781 // At each iteration, move all visitors from report to visitor list. This is
2782 // important, because the Profile() functions of the visitors make sure that
2783 // a visitor isn't added multiple times for the same node, but it's fine
2784 // to add the a visitor with Profile() for different nodes (e.g. tracking
2785 // a region at different points of the symbolic execution).
2786 for (std::unique_ptr<BugReporterVisitor> &Visitor : R->visitors())
2787 visitors.push_back(std::move(Visitor));
2788
2789 R->clearVisitors();
2790
2791 const ExplodedNode *Pred = NextNode->getFirstPred();
2792 if (!Pred) {
2793 PathDiagnosticPieceRef LastPiece;
2794 for (auto &V : visitors) {
2795 V->finalizeVisitor(BRC, ErrorNode, *R);
2796
2797 if (auto Piece = V->getEndPath(BRC, ErrorNode, *R)) {
2798 assert(!LastPiece &&
2799 "There can only be one final piece in a diagnostic.");
2800 assert(Piece->getKind() == PathDiagnosticPiece::Kind::Event &&
2801 "The final piece must contain a message!");
2802 LastPiece = std::move(Piece);
2803 (*Notes)[ErrorNode].push_back(LastPiece);
2804 }
2805 }
2806 break;
2807 }
2808
2809 for (auto &V : visitors) {
2810 auto P = V->VisitNode(NextNode, BRC, *R);
2811 if (P)
2812 (*Notes)[NextNode].push_back(std::move(P));
2813 }
2814
2815 if (!R->isValid())
2816 break;
2817
2818 NextNode = Pred;
2819 }
2820
2821 return Notes;
2822}
2823
2824Optional<PathDiagnosticBuilder> PathDiagnosticBuilder::findValidReport(
2825 ArrayRef<PathSensitiveBugReport *> &bugReports,
2826 PathSensitiveBugReporter &Reporter) {
2827
2828 BugPathGetter BugGraph(&Reporter.getGraph(), bugReports);
2829
2830 while (BugPathInfo *BugPath = BugGraph.getNextBugPath()) {
2831 // Find the BugReport with the original location.
2832 PathSensitiveBugReport *R = BugPath->Report;
2833 assert(R && "No original report found for sliced graph.");
2834 assert(R->isValid() && "Report selected by trimmed graph marked invalid.");
2835 const ExplodedNode *ErrorNode = BugPath->ErrorNode;
2836
2837 // Register refutation visitors first, if they mark the bug invalid no
2838 // further analysis is required
2839 R->addVisitor<LikelyFalsePositiveSuppressionBRVisitor>();
2840
2841 // Register additional node visitors.
2842 R->addVisitor<NilReceiverBRVisitor>();
2843 R->addVisitor<ConditionBRVisitor>();
2844 R->addVisitor<TagVisitor>();
2845
2846 BugReporterContext BRC(Reporter);
2847
2848 // Run all visitors on a given graph, once.
2849 std::unique_ptr<VisitorsDiagnosticsTy> visitorNotes =
2850 generateVisitorsDiagnostics(R, ErrorNode, BRC);
2851
2852 if (R->isValid()) {
2853 if (Reporter.getAnalyzerOptions().ShouldCrosscheckWithZ3) {
2854 // If crosscheck is enabled, remove all visitors, add the refutation
2855 // visitor and check again
2856 R->clearVisitors();
2857 R->addVisitor<FalsePositiveRefutationBRVisitor>();
2858
2859 // We don't overwrite the notes inserted by other visitors because the
2860 // refutation manager does not add any new note to the path
2861 generateVisitorsDiagnostics(R, BugPath->ErrorNode, BRC);
2862 }
2863
2864 // Check if the bug is still valid
2865 if (R->isValid())
2866 return PathDiagnosticBuilder(
2867 std::move(BRC), std::move(BugPath->BugPath), BugPath->Report,
2868 BugPath->ErrorNode, std::move(visitorNotes));
2869 }
2870 }
2871
2872 return {};
2873}
2874
2875std::unique_ptr<DiagnosticForConsumerMapTy>
2876PathSensitiveBugReporter::generatePathDiagnostics(
2877 ArrayRef<PathDiagnosticConsumer *> consumers,
2878 ArrayRef<PathSensitiveBugReport *> &bugReports) {
2879 assert(!bugReports.empty());
2880
2881 auto Out = std::make_unique<DiagnosticForConsumerMapTy>();
2882
2883 Optional<PathDiagnosticBuilder> PDB =
2884 PathDiagnosticBuilder::findValidReport(bugReports, *this);
2885
2886 if (PDB) {
2887 for (PathDiagnosticConsumer *PC : consumers) {
2888 if (std::unique_ptr<PathDiagnostic> PD = PDB->generate(PC)) {
2889 (*Out)[PC] = std::move(PD);
2890 }
2891 }
2892 }
2893
2894 return Out;
2895}
2896
2897void BugReporter::emitReport(std::unique_ptr<BugReport> R) {
2898 bool ValidSourceLoc = R->getLocation().isValid();
2899 assert(ValidSourceLoc);
2900 // If we mess up in a release build, we'd still prefer to just drop the bug
2901 // instead of trying to go on.
2902 if (!ValidSourceLoc)
2903 return;
2904
2905 // Compute the bug report's hash to determine its equivalence class.
2906 llvm::FoldingSetNodeID ID;
2907 R->Profile(ID);
2908
2909 // Lookup the equivance class. If there isn't one, create it.
2910 void *InsertPos;
2911 BugReportEquivClass* EQ = EQClasses.FindNodeOrInsertPos(ID, InsertPos);
2912
2913 if (!EQ) {
2914 EQ = new BugReportEquivClass(std::move(R));
2915 EQClasses.InsertNode(EQ, InsertPos);
2916 EQClassesVector.push_back(EQ);
2917 } else
2918 EQ->AddReport(std::move(R));
2919}
2920
2921void PathSensitiveBugReporter::emitReport(std::unique_ptr<BugReport> R) {
2922 if (auto PR = dyn_cast<PathSensitiveBugReport>(R.get()))
2923 if (const ExplodedNode *E = PR->getErrorNode()) {
2924 // An error node must either be a sink or have a tag, otherwise
2925 // it could get reclaimed before the path diagnostic is created.
2926 assert((E->isSink() || E->getLocation().getTag()) &&
2927 "Error node must either be a sink or have a tag");
2928
2929 const AnalysisDeclContext *DeclCtx =
2930 E->getLocationContext()->getAnalysisDeclContext();
2931 // The source of autosynthesized body can be handcrafted AST or a model
2932 // file. The locations from handcrafted ASTs have no valid source
2933 // locations and have to be discarded. Locations from model files should
2934 // be preserved for processing and reporting.
2935 if (DeclCtx->isBodyAutosynthesized() &&
2936 !DeclCtx->isBodyAutosynthesizedFromModelFile())
2937 return;
2938 }
2939
2940 BugReporter::emitReport(std::move(R));
2941}
2942
2943//===----------------------------------------------------------------------===//
2944// Emitting reports in equivalence classes.
2945//===----------------------------------------------------------------------===//
2946
2947namespace {
2948
2949struct FRIEC_WLItem {
2950 const ExplodedNode *N;
2951 ExplodedNode::const_succ_iterator I, E;
2952
2953 FRIEC_WLItem(const ExplodedNode *n)
2954 : N(n), I(N->succ_begin()), E(N->succ_end()) {}
2955};
2956
2957} // namespace
2958
2959BugReport *PathSensitiveBugReporter::findReportInEquivalenceClass(
2960 BugReportEquivClass &EQ, SmallVectorImpl<BugReport *> &bugReports) {
2961 // If we don't need to suppress any of the nodes because they are
2962 // post-dominated by a sink, simply add all the nodes in the equivalence class
2963 // to 'Nodes'. Any of the reports will serve as a "representative" report.
2964 assert(EQ.getReports().size() > 0);
2965 const BugType& BT = EQ.getReports()[0]->getBugType();
2966 if (!BT.isSuppressOnSink()) {
2967 BugReport *R = EQ.getReports()[0].get();
2968 for (auto &J : EQ.getReports()) {
2969 if (auto *PR = dyn_cast<PathSensitiveBugReport>(J.get())) {
2970 R = PR;
2971 bugReports.push_back(PR);
2972 }
2973 }
2974 return R;
2975 }
2976
2977 // For bug reports that should be suppressed when all paths are post-dominated
2978 // by a sink node, iterate through the reports in the equivalence class
2979 // until we find one that isn't post-dominated (if one exists). We use a
2980 // DFS traversal of the ExplodedGraph to find a non-sink node. We could write
2981 // this as a recursive function, but we don't want to risk blowing out the
2982 // stack for very long paths.
2983 BugReport *exampleReport = nullptr;
2984
2985 for (const auto &I: EQ.getReports()) {
2986 auto *R = dyn_cast<PathSensitiveBugReport>(I.get());
2987 if (!R)
2988 continue;
2989
2990 const ExplodedNode *errorNode = R->getErrorNode();
2991 if (errorNode->isSink()) {
2992 llvm_unreachable(
2993 "BugType::isSuppressSink() should not be 'true' for sink end nodes");
2994 }
2995 // No successors? By definition this nodes isn't post-dominated by a sink.
2996 if (errorNode->succ_empty()) {
2997 bugReports.push_back(R);
2998 if (!exampleReport)
2999 exampleReport = R;
3000 continue;
3001 }
3002
3003 // See if we are in a no-return CFG block. If so, treat this similarly
3004 // to being post-dominated by a sink. This works better when the analysis
3005 // is incomplete and we have never reached the no-return function call(s)
3006 // that we'd inevitably bump into on this path.
3007 if (const CFGBlock *ErrorB = errorNode->getCFGBlock())
3008 if (ErrorB->isInevitablySinking())
3009 continue;
3010
3011 // At this point we know that 'N' is not a sink and it has at least one
3012 // successor. Use a DFS worklist to find a non-sink end-of-path node.
3013 using WLItem = FRIEC_WLItem;
3014 using DFSWorkList = SmallVector<WLItem, 10>;
3015
3016 llvm::DenseMap<const ExplodedNode *, unsigned> Visited;
3017
3018 DFSWorkList WL;
3019 WL.push_back(errorNode);
3020 Visited[errorNode] = 1;
3021
3022 while (!WL.empty()) {
3023 WLItem &WI = WL.back();
3024 assert(!WI.N->succ_empty());
3025
3026 for (; WI.I != WI.E; ++WI.I) {
3027 const ExplodedNode *Succ = *WI.I;
3028 // End-of-path node?
3029 if (Succ->succ_empty()) {
3030 // If we found an end-of-path node that is not a sink.
3031 if (!Succ->isSink()) {
3032 bugReports.push_back(R);
3033 if (!exampleReport)
3034 exampleReport = R;
3035 WL.clear();
3036 break;
3037 }
3038 // Found a sink? Continue on to the next successor.
3039 continue;
3040 }
3041 // Mark the successor as visited. If it hasn't been explored,
3042 // enqueue it to the DFS worklist.
3043 unsigned &mark = Visited[Succ];
3044 if (!mark) {
3045 mark = 1;
3046 WL.push_back(Succ);
3047 break;
3048 }
3049 }
3050
3051 // The worklist may have been cleared at this point. First
3052 // check if it is empty before checking the last item.
3053 if (!WL.empty() && &WL.back() == &WI)
3054 WL.pop_back();
3055 }
3056 }
3057
3058 // ExampleReport will be NULL if all the nodes in the equivalence class
3059 // were post-dominated by sinks.
3060 return exampleReport;
3061}
3062
3063void BugReporter::FlushReport(BugReportEquivClass& EQ) {
3064 SmallVector<BugReport*, 10> bugReports;
3065 BugReport *report = findReportInEquivalenceClass(EQ, bugReports);
3066 if (!report)
3067 return;
3068
3069 // See whether we need to silence the checker/package.
3070 for (const std::string &CheckerOrPackage :
3071 getAnalyzerOptions().SilencedCheckersAndPackages) {
3072 if (report->getBugType().getCheckerName().startswith(
3073 CheckerOrPackage))
3074 return;
3075 }
3076
3077 ArrayRef<PathDiagnosticConsumer*> Consumers = getPathDiagnosticConsumers();
3078 std::unique_ptr<DiagnosticForConsumerMapTy> Diagnostics =
3079 generateDiagnosticForConsumerMap(report, Consumers, bugReports);
3080
3081 for (auto &P : *Diagnostics) {
3082 PathDiagnosticConsumer *Consumer = P.first;
3083 std::unique_ptr<PathDiagnostic> &PD = P.second;
3084
3085 // If the path is empty, generate a single step path with the location
3086 // of the issue.
3087 if (PD->path.empty()) {
3088 PathDiagnosticLocation L = report->getLocation();
3089 auto piece = std::make_unique<PathDiagnosticEventPiece>(
3090 L, report->getDescription());
3091 for (SourceRange Range : report->getRanges())
3092 piece->addRange(Range);
3093 PD->setEndOfPath(std::move(piece));
3094 }
3095
3096 PathPieces &Pieces = PD->getMutablePieces();
3097 if (getAnalyzerOptions().ShouldDisplayNotesAsEvents) {
3098 // For path diagnostic consumers that don't support extra notes,
3099 // we may optionally convert those to path notes.
3100 for (auto I = report->getNotes().rbegin(),
3101 E = report->getNotes().rend(); I != E; ++I) {
3102 PathDiagnosticNotePiece *Piece = I->get();
3103 auto ConvertedPiece = std::make_shared<PathDiagnosticEventPiece>(
3104 Piece->getLocation(), Piece->getString());
3105 for (const auto &R: Piece->getRanges())
3106 ConvertedPiece->addRange(R);
3107
3108 Pieces.push_front(std::move(ConvertedPiece));
3109 }
3110 } else {
3111 for (auto I = report->getNotes().rbegin(),
3112 E = report->getNotes().rend(); I != E; ++I)
3113 Pieces.push_front(*I);
3114 }
3115
3116 for (const auto &I : report->getFixits())
3117 Pieces.back()->addFixit(I);
3118
3119 updateExecutedLinesWithDiagnosticPieces(*PD);
3120 Consumer->HandlePathDiagnostic(std::move(PD));
3121 }
3122}
3123
3124/// Insert all lines participating in the function signature \p Signature
3125/// into \p ExecutedLines.
3126static void populateExecutedLinesWithFunctionSignature(
3127 const Decl *Signature, const SourceManager &SM,
3128 FilesToLineNumsMap &ExecutedLines) {
3129 SourceRange SignatureSourceRange;
3130 const Stmt* Body = Signature->getBody();
3131 if (const auto FD = dyn_cast<FunctionDecl>(Signature)) {
3132 SignatureSourceRange = FD->getSourceRange();
3133 } else if (const auto OD = dyn_cast<ObjCMethodDecl>(Signature)) {
3134 SignatureSourceRange = OD->getSourceRange();
3135 } else {
3136 return;
3137 }
3138 SourceLocation Start = SignatureSourceRange.getBegin();
3139 SourceLocation End = Body ? Body->getSourceRange().getBegin()
3140 : SignatureSourceRange.getEnd();
3141 if (!Start.isValid() || !End.isValid())
3142 return;
3143 unsigned StartLine = SM.getExpansionLineNumber(Start);
3144 unsigned EndLine = SM.getExpansionLineNumber(End);
3145
3146 FileID FID = SM.getFileID(SM.getExpansionLoc(Start));
3147 for (unsigned Line = StartLine; Line <= EndLine; Line++)
3148 ExecutedLines[FID].insert(Line);
3149}
3150
3151static void populateExecutedLinesWithStmt(
3152 const Stmt *S, const SourceManager &SM,
3153 FilesToLineNumsMap &ExecutedLines) {
3154 SourceLocation Loc = S->getSourceRange().getBegin();
3155 if (!Loc.isValid())
3156 return;
3157 SourceLocation ExpansionLoc = SM.getExpansionLoc(Loc);
3158 FileID FID = SM.getFileID(ExpansionLoc);
3159 unsigned LineNo = SM.getExpansionLineNumber(ExpansionLoc);
3160 ExecutedLines[FID].insert(LineNo);
3161}
3162
3163/// \return all executed lines including function signatures on the path
3164/// starting from \p N.
3165static std::unique_ptr<FilesToLineNumsMap>
3166findExecutedLines(const SourceManager &SM, const ExplodedNode *N) {
3167 auto ExecutedLines = std::make_unique<FilesToLineNumsMap>();
3168
3169 while (N) {
3170 if (N->getFirstPred() == nullptr) {
3171 // First node: show signature of the entrance point.
3172 const Decl *D = N->getLocationContext()->getDecl();
3173 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines);
3174 } else if (auto CE = N->getLocationAs<CallEnter>()) {
3175 // Inlined function: show signature.
3176 const Decl* D = CE->getCalleeContext()->getDecl();
3177 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines);
3178 } else if (const Stmt *S = N->getStmtForDiagnostics()) {
3179 populateExecutedLinesWithStmt(S, SM, *ExecutedLines);
3180
3181 // Show extra context for some parent kinds.
3182 const Stmt *P = N->getParentMap().getParent(S);
3183
3184 // The path exploration can die before the node with the associated
3185 // return statement is generated, but we do want to show the whole
3186 // return.
3187 if (const auto *RS = dyn_cast_or_null<ReturnStmt>(P)) {
3188 populateExecutedLinesWithStmt(RS, SM, *ExecutedLines);
3189 P = N->getParentMap().getParent(RS);
3190 }
3191
3192 if (isa_and_nonnull<SwitchCase, LabelStmt>(P))
3193 populateExecutedLinesWithStmt(P, SM, *ExecutedLines);
3194 }
3195
3196 N = N->getFirstPred();
3197 }
3198 return ExecutedLines;
3199}
3200
3201std::unique_ptr<DiagnosticForConsumerMapTy>
3202BugReporter::generateDiagnosticForConsumerMap(
3203 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers,
3204 ArrayRef<BugReport *> bugReports) {
3205 auto *basicReport = cast<BasicBugReport>(exampleReport);
3206 auto Out = std::make_unique<DiagnosticForConsumerMapTy>();
3207 for (auto *Consumer : consumers)
3208 (*Out)[Consumer] = generateDiagnosticForBasicReport(basicReport);
3209 return Out;
3210}
3211
3212static PathDiagnosticCallPiece *
3213getFirstStackedCallToHeaderFile(PathDiagnosticCallPiece *CP,
3214 const SourceManager &SMgr) {
3215 SourceLocation CallLoc = CP->callEnter.asLocation();
3216
3217 // If the call is within a macro, don't do anything (for now).
3218 if (CallLoc.isMacroID())
3219 return nullptr;
3220
3221 assert(AnalysisManager::isInCodeFile(CallLoc, SMgr) &&
3222 "The call piece should not be in a header file.");
3223
3224 // Check if CP represents a path through a function outside of the main file.
3225 if (!AnalysisManager::isInCodeFile(CP->callEnterWithin.asLocation(), SMgr))
3226 return CP;
3227
3228 const PathPieces &Path = CP->path;
3229 if (Path.empty())
3230 return nullptr;
3231
3232 // Check if the last piece in the callee path is a call to a function outside
3233 // of the main file.
3234 if (auto *CPInner = dyn_cast<PathDiagnosticCallPiece>(Path.back().get()))
3235 return getFirstStackedCallToHeaderFile(CPInner, SMgr);
3236
3237 // Otherwise, the last piece is in the main file.
3238 return nullptr;
3239}
3240
3241static void resetDiagnosticLocationToMainFile(PathDiagnostic &PD) {
3242 if (PD.path.empty())
3243 return;
3244
3245 PathDiagnosticPiece *LastP = PD.path.back().get();
3246 assert(LastP);
3247 const SourceManager &SMgr = LastP->getLocation().getManager();
3248
3249 // We only need to check if the report ends inside headers, if the last piece
3250 // is a call piece.
3251 if (auto *CP = dyn_cast<PathDiagnosticCallPiece>(LastP)) {
3252 CP = getFirstStackedCallToHeaderFile(CP, SMgr);
3253 if (CP) {
3254 // Mark the piece.
3255 CP->setAsLastInMainSourceFile();
3256
3257 // Update the path diagnostic message.
3258 const auto *ND = dyn_cast<NamedDecl>(CP->getCallee());
3259 if (ND) {
3260 SmallString<200> buf;
3261 llvm::raw_svector_ostream os(buf);
3262 os << " (within a call to '" << ND->getDeclName() << "')";
3263 PD.appendToDesc(os.str());
3264 }
3265
3266 // Reset the report containing declaration and location.
3267 PD.setDeclWithIssue(CP->getCaller());
3268 PD.setLocation(CP->getLocation());
3269
3270 return;
3271 }
3272 }
3273}
3274
3275
3276
3277std::unique_ptr<DiagnosticForConsumerMapTy>
3278PathSensitiveBugReporter::generateDiagnosticForConsumerMap(
3279 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers,
3280 ArrayRef<BugReport *> bugReports) {
3281 std::vector<BasicBugReport *> BasicBugReports;
3282 std::vector<PathSensitiveBugReport *> PathSensitiveBugReports;
3283 if (isa<BasicBugReport>(exampleReport))
3284 return BugReporter::generateDiagnosticForConsumerMap(exampleReport,
3285 consumers, bugReports);
3286
3287 // Generate the full path sensitive diagnostic, using the generation scheme
3288 // specified by the PathDiagnosticConsumer. Note that we have to generate
3289 // path diagnostics even for consumers which do not support paths, because
3290 // the BugReporterVisitors may mark this bug as a false positive.
3291 assert(!bugReports.empty());
3292 MaxBugClassSize.updateMax(bugReports.size());
3293
3294 // Avoid copying the whole array because there may be a lot of reports.
3295 ArrayRef<PathSensitiveBugReport *> convertedArrayOfReports(
3296 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.begin()),
3297 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.end()));
3298 std::unique_ptr<DiagnosticForConsumerMapTy> Out = generatePathDiagnostics(
3299 consumers, convertedArrayOfReports);
3300
3301 if (Out->empty())
3302 return Out;
3303
3304 MaxValidBugClassSize.updateMax(bugReports.size());
3305
3306 // Examine the report and see if the last piece is in a header. Reset the
3307 // report location to the last piece in the main source file.
3308 const AnalyzerOptions &Opts = getAnalyzerOptions();
3309 for (auto const &P : *Out)
3310 if (Opts.ShouldReportIssuesInMainSourceFile && !Opts.AnalyzeAll)
3311 resetDiagnosticLocationToMainFile(*P.second);
3312
3313 return Out;
3314}
3315
3316void BugReporter::EmitBasicReport(const Decl *DeclWithIssue,
3317 const CheckerBase *Checker, StringRef Name,
3318 StringRef Category, StringRef Str,
3319 PathDiagnosticLocation Loc,
3320 ArrayRef<SourceRange> Ranges,
3321 ArrayRef<FixItHint> Fixits) {
3322 EmitBasicReport(DeclWithIssue, Checker->getCheckerName(), Name, Category, Str,
3323 Loc, Ranges, Fixits);
3324}
3325
3326void BugReporter::EmitBasicReport(const Decl *DeclWithIssue,
3327 CheckerNameRef CheckName,
3328 StringRef name, StringRef category,
3329 StringRef str, PathDiagnosticLocation Loc,
3330 ArrayRef<SourceRange> Ranges,
3331 ArrayRef<FixItHint> Fixits) {
3332 // 'BT' is owned by BugReporter.
3333 BugType *BT = getBugTypeForName(CheckName, name, category);
3334 auto R = std::make_unique<BasicBugReport>(*BT, str, Loc);
3335 R->setDeclWithIssue(DeclWithIssue);
3336 for (const auto &SR : Ranges)
3337 R->addRange(SR);
3338 for (const auto &FH : Fixits)
3339 R->addFixItHint(FH);
3340 emitReport(std::move(R));
3341}
3342
3343BugType *BugReporter::getBugTypeForName(CheckerNameRef CheckName,
3344 StringRef name, StringRef category) {
3345 SmallString<136> fullDesc;
3346 llvm::raw_svector_ostream(fullDesc) << CheckName.getName() << ":" << name
3347 << ":" << category;
3348 std::unique_ptr<BugType> &BT = StrBugTypes[fullDesc];
3349 if (!BT)
3350 BT = std::make_unique<BugType>(CheckName, name, category);
3351 return BT.get();
3352}
3353

source code of clang/lib/StaticAnalyzer/Core/BugReporter.cpp