1 //===--- LoopConvertUtils.cpp - clang-tidy --------------------------------===//
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 #include "LoopConvertUtils.h"
10 #include "clang/Basic/IdentifierTable.h"
11 #include "clang/Basic/LLVM.h"
12 #include "clang/Basic/Lambda.h"
13 #include "clang/Basic/SourceLocation.h"
14 #include "clang/Basic/SourceManager.h"
15 #include "clang/Basic/TokenKinds.h"
16 #include "clang/Lex/Lexer.h"
17 #include "llvm/ADT/APSInt.h"
18 #include "llvm/ADT/FoldingSet.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/Support/Casting.h"
21 #include <algorithm>
22 #include <cassert>
23 #include <cstddef>
24 #include <string>
25 #include <utility>
26 
27 using namespace clang::ast_matchers;
28 
29 namespace clang {
30 namespace tidy {
31 namespace modernize {
32 
33 /// Tracks a stack of parent statements during traversal.
34 ///
35 /// All this really does is inject push_back() before running
36 /// RecursiveASTVisitor::TraverseStmt() and pop_back() afterwards. The Stmt atop
37 /// the stack is the parent of the current statement (NULL for the topmost
38 /// statement).
TraverseStmt(Stmt * Statement)39 bool StmtAncestorASTVisitor::TraverseStmt(Stmt *Statement) {
40   StmtAncestors.insert(std::make_pair(Statement, StmtStack.back()));
41   StmtStack.push_back(Statement);
42   RecursiveASTVisitor<StmtAncestorASTVisitor>::TraverseStmt(Statement);
43   StmtStack.pop_back();
44   return true;
45 }
46 
47 /// Keep track of the DeclStmt associated with each VarDecl.
48 ///
49 /// Combined with StmtAncestors, this provides roughly the same information as
50 /// Scope, as we can map a VarDecl to its DeclStmt, then walk up the parent tree
51 /// using StmtAncestors.
VisitDeclStmt(DeclStmt * Decls)52 bool StmtAncestorASTVisitor::VisitDeclStmt(DeclStmt *Decls) {
53   for (const auto *Decl : Decls->decls()) {
54     if (const auto *V = dyn_cast<VarDecl>(Decl))
55       DeclParents.insert(std::make_pair(V, Decls));
56   }
57   return true;
58 }
59 
60 /// record the DeclRefExpr as part of the parent expression.
VisitDeclRefExpr(DeclRefExpr * E)61 bool ComponentFinderASTVisitor::VisitDeclRefExpr(DeclRefExpr *E) {
62   Components.push_back(E);
63   return true;
64 }
65 
66 /// record the MemberExpr as part of the parent expression.
VisitMemberExpr(MemberExpr * Member)67 bool ComponentFinderASTVisitor::VisitMemberExpr(MemberExpr *Member) {
68   Components.push_back(Member);
69   return true;
70 }
71 
72 /// Forward any DeclRefExprs to a check on the referenced variable
73 /// declaration.
VisitDeclRefExpr(DeclRefExpr * DeclRef)74 bool DependencyFinderASTVisitor::VisitDeclRefExpr(DeclRefExpr *DeclRef) {
75   if (auto *V = dyn_cast_or_null<VarDecl>(DeclRef->getDecl()))
76     return VisitVarDecl(V);
77   return true;
78 }
79 
80 /// Determine if any this variable is declared inside the ContainingStmt.
VisitVarDecl(VarDecl * V)81 bool DependencyFinderASTVisitor::VisitVarDecl(VarDecl *V) {
82   const Stmt *Curr = DeclParents->lookup(V);
83   // First, see if the variable was declared within an inner scope of the loop.
84   while (Curr != nullptr) {
85     if (Curr == ContainingStmt) {
86       DependsOnInsideVariable = true;
87       return false;
88     }
89     Curr = StmtParents->lookup(Curr);
90   }
91 
92   // Next, check if the variable was removed from existence by an earlier
93   // iteration.
94   for (const auto &I : *ReplacedVars) {
95     if (I.second == V) {
96       DependsOnInsideVariable = true;
97       return false;
98     }
99   }
100   return true;
101 }
102 
103 /// If we already created a variable for TheLoop, check to make sure
104 /// that the name was not already taken.
VisitForStmt(ForStmt * TheLoop)105 bool DeclFinderASTVisitor::VisitForStmt(ForStmt *TheLoop) {
106   StmtGeneratedVarNameMap::const_iterator I = GeneratedDecls->find(TheLoop);
107   if (I != GeneratedDecls->end() && I->second == Name) {
108     Found = true;
109     return false;
110   }
111   return true;
112 }
113 
114 /// If any named declaration within the AST subtree has the same name,
115 /// then consider Name already taken.
VisitNamedDecl(NamedDecl * D)116 bool DeclFinderASTVisitor::VisitNamedDecl(NamedDecl *D) {
117   const IdentifierInfo *Ident = D->getIdentifier();
118   if (Ident && Ident->getName() == Name) {
119     Found = true;
120     return false;
121   }
122   return true;
123 }
124 
125 /// Forward any declaration references to the actual check on the
126 /// referenced declaration.
VisitDeclRefExpr(DeclRefExpr * DeclRef)127 bool DeclFinderASTVisitor::VisitDeclRefExpr(DeclRefExpr *DeclRef) {
128   if (auto *D = dyn_cast<NamedDecl>(DeclRef->getDecl()))
129     return VisitNamedDecl(D);
130   return true;
131 }
132 
133 /// If the new variable name conflicts with any type used in the loop,
134 /// then we mark that variable name as taken.
VisitTypeLoc(TypeLoc TL)135 bool DeclFinderASTVisitor::VisitTypeLoc(TypeLoc TL) {
136   QualType QType = TL.getType();
137 
138   // Check if our name conflicts with a type, to handle for typedefs.
139   if (QType.getAsString() == Name) {
140     Found = true;
141     return false;
142   }
143   // Check for base type conflicts. For example, when a struct is being
144   // referenced in the body of the loop, the above getAsString() will return the
145   // whole type (ex. "struct s"), but will be caught here.
146   if (const IdentifierInfo *Ident = QType.getBaseTypeIdentifier()) {
147     if (Ident->getName() == Name) {
148       Found = true;
149       return false;
150     }
151   }
152   return true;
153 }
154 
155 /// Look through conversion/copy constructors to find the explicit
156 /// initialization expression, returning it is found.
157 ///
158 /// The main idea is that given
159 ///   vector<int> v;
160 /// we consider either of these initializations
161 ///   vector<int>::iterator it = v.begin();
162 ///   vector<int>::iterator it(v.begin());
163 /// and retrieve `v.begin()` as the expression used to initialize `it` but do
164 /// not include
165 ///   vector<int>::iterator it;
166 ///   vector<int>::iterator it(v.begin(), 0); // if this constructor existed
167 /// as being initialized from `v.begin()`
digThroughConstructors(const Expr * E)168 const Expr *digThroughConstructors(const Expr *E) {
169   if (!E)
170     return nullptr;
171   E = E->IgnoreImplicit();
172   if (const auto *ConstructExpr = dyn_cast<CXXConstructExpr>(E)) {
173     // The initial constructor must take exactly one parameter, but base class
174     // and deferred constructors can take more.
175     if (ConstructExpr->getNumArgs() != 1 ||
176         ConstructExpr->getConstructionKind() != CXXConstructExpr::CK_Complete)
177       return nullptr;
178     E = ConstructExpr->getArg(0);
179     if (const auto *Temp = dyn_cast<MaterializeTemporaryExpr>(E))
180       E = Temp->getSubExpr();
181     return digThroughConstructors(E);
182   }
183   return E;
184 }
185 
186 /// Returns true when two Exprs are equivalent.
areSameExpr(ASTContext * Context,const Expr * First,const Expr * Second)187 bool areSameExpr(ASTContext *Context, const Expr *First, const Expr *Second) {
188   if (!First || !Second)
189     return false;
190 
191   llvm::FoldingSetNodeID FirstID, SecondID;
192   First->Profile(FirstID, *Context, true);
193   Second->Profile(SecondID, *Context, true);
194   return FirstID == SecondID;
195 }
196 
197 /// Returns the DeclRefExpr represented by E, or NULL if there isn't one.
getDeclRef(const Expr * E)198 const DeclRefExpr *getDeclRef(const Expr *E) {
199   return dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts());
200 }
201 
202 /// Returns true when two ValueDecls are the same variable.
areSameVariable(const ValueDecl * First,const ValueDecl * Second)203 bool areSameVariable(const ValueDecl *First, const ValueDecl *Second) {
204   return First && Second &&
205          First->getCanonicalDecl() == Second->getCanonicalDecl();
206 }
207 
208 /// Determines if an expression is a declaration reference to a
209 /// particular variable.
exprReferencesVariable(const ValueDecl * Target,const Expr * E)210 static bool exprReferencesVariable(const ValueDecl *Target, const Expr *E) {
211   if (!Target || !E)
212     return false;
213   const DeclRefExpr *Decl = getDeclRef(E);
214   return Decl && areSameVariable(Target, Decl->getDecl());
215 }
216 
217 /// If the expression is a dereference or call to operator*(), return the
218 /// operand. Otherwise, return NULL.
getDereferenceOperand(const Expr * E)219 static const Expr *getDereferenceOperand(const Expr *E) {
220   if (const auto *Uop = dyn_cast<UnaryOperator>(E))
221     return Uop->getOpcode() == UO_Deref ? Uop->getSubExpr() : nullptr;
222 
223   if (const auto *OpCall = dyn_cast<CXXOperatorCallExpr>(E)) {
224     return OpCall->getOperator() == OO_Star && OpCall->getNumArgs() == 1
225                ? OpCall->getArg(0)
226                : nullptr;
227   }
228 
229   return nullptr;
230 }
231 
232 /// Returns true when the Container contains an Expr equivalent to E.
233 template <typename ContainerT>
containsExpr(ASTContext * Context,const ContainerT * Container,const Expr * E)234 static bool containsExpr(ASTContext *Context, const ContainerT *Container,
235                          const Expr *E) {
236   llvm::FoldingSetNodeID ID;
237   E->Profile(ID, *Context, true);
238   for (const auto &I : *Container) {
239     if (ID == I.second)
240       return true;
241   }
242   return false;
243 }
244 
245 /// Returns true when the index expression is a declaration reference to
246 /// IndexVar.
247 ///
248 /// If the index variable is `index`, this function returns true on
249 ///    arrayExpression[index];
250 ///    containerExpression[index];
251 /// but not
252 ///    containerExpression[notIndex];
isIndexInSubscriptExpr(const Expr * IndexExpr,const VarDecl * IndexVar)253 static bool isIndexInSubscriptExpr(const Expr *IndexExpr,
254                                    const VarDecl *IndexVar) {
255   const DeclRefExpr *Idx = getDeclRef(IndexExpr);
256   return Idx && Idx->getType()->isIntegerType() &&
257          areSameVariable(IndexVar, Idx->getDecl());
258 }
259 
260 /// Returns true when the index expression is a declaration reference to
261 /// IndexVar, Obj is the same expression as SourceExpr after all parens and
262 /// implicit casts are stripped off.
263 ///
264 /// If PermitDeref is true, IndexExpression may
265 /// be a dereference (overloaded or builtin operator*).
266 ///
267 /// This function is intended for array-like containers, as it makes sure that
268 /// both the container and the index match.
269 /// If the loop has index variable `index` and iterates over `container`, then
270 /// isIndexInSubscriptExpr returns true for
271 /// \code
272 ///   container[index]
273 ///   container.at(index)
274 ///   container->at(index)
275 /// \endcode
276 /// but not for
277 /// \code
278 ///   container[notIndex]
279 ///   notContainer[index]
280 /// \endcode
281 /// If PermitDeref is true, then isIndexInSubscriptExpr additionally returns
282 /// true on these expressions:
283 /// \code
284 ///   (*container)[index]
285 ///   (*container).at(index)
286 /// \endcode
isIndexInSubscriptExpr(ASTContext * Context,const Expr * IndexExpr,const VarDecl * IndexVar,const Expr * Obj,const Expr * SourceExpr,bool PermitDeref)287 static bool isIndexInSubscriptExpr(ASTContext *Context, const Expr *IndexExpr,
288                                    const VarDecl *IndexVar, const Expr *Obj,
289                                    const Expr *SourceExpr, bool PermitDeref) {
290   if (!SourceExpr || !Obj || !isIndexInSubscriptExpr(IndexExpr, IndexVar))
291     return false;
292 
293   if (areSameExpr(Context, SourceExpr->IgnoreParenImpCasts(),
294                   Obj->IgnoreParenImpCasts()))
295     return true;
296 
297   if (const Expr *InnerObj = getDereferenceOperand(Obj->IgnoreParenImpCasts()))
298     if (PermitDeref && areSameExpr(Context, SourceExpr->IgnoreParenImpCasts(),
299                                    InnerObj->IgnoreParenImpCasts()))
300       return true;
301 
302   return false;
303 }
304 
305 /// Returns true when Opcall is a call a one-parameter dereference of
306 /// IndexVar.
307 ///
308 /// For example, if the index variable is `index`, returns true for
309 ///   *index
310 /// but not
311 ///   index
312 ///   *notIndex
isDereferenceOfOpCall(const CXXOperatorCallExpr * OpCall,const VarDecl * IndexVar)313 static bool isDereferenceOfOpCall(const CXXOperatorCallExpr *OpCall,
314                                   const VarDecl *IndexVar) {
315   return OpCall->getOperator() == OO_Star && OpCall->getNumArgs() == 1 &&
316          exprReferencesVariable(IndexVar, OpCall->getArg(0));
317 }
318 
319 /// Returns true when Uop is a dereference of IndexVar.
320 ///
321 /// For example, if the index variable is `index`, returns true for
322 ///   *index
323 /// but not
324 ///   index
325 ///   *notIndex
isDereferenceOfUop(const UnaryOperator * Uop,const VarDecl * IndexVar)326 static bool isDereferenceOfUop(const UnaryOperator *Uop,
327                                const VarDecl *IndexVar) {
328   return Uop->getOpcode() == UO_Deref &&
329          exprReferencesVariable(IndexVar, Uop->getSubExpr());
330 }
331 
332 /// Determines whether the given Decl defines a variable initialized to
333 /// the loop object.
334 ///
335 /// This is intended to find cases such as
336 /// \code
337 ///   for (int i = 0; i < arraySize(arr); ++i) {
338 ///     T t = arr[i];
339 ///     // use t, do not use i
340 ///   }
341 /// \endcode
342 /// and
343 /// \code
344 ///   for (iterator i = container.begin(), e = container.end(); i != e; ++i) {
345 ///     T t = *i;
346 ///     // use t, do not use i
347 ///   }
348 /// \endcode
isAliasDecl(ASTContext * Context,const Decl * TheDecl,const VarDecl * IndexVar)349 static bool isAliasDecl(ASTContext *Context, const Decl *TheDecl,
350                         const VarDecl *IndexVar) {
351   const auto *VDecl = dyn_cast<VarDecl>(TheDecl);
352   if (!VDecl)
353     return false;
354   if (!VDecl->hasInit())
355     return false;
356 
357   bool OnlyCasts = true;
358   const Expr *Init = VDecl->getInit()->IgnoreParenImpCasts();
359   if (isa_and_nonnull<CXXConstructExpr>(Init)) {
360     Init = digThroughConstructors(Init);
361     OnlyCasts = false;
362   }
363   if (!Init)
364     return false;
365 
366   // Check that the declared type is the same as (or a reference to) the
367   // container type.
368   if (!OnlyCasts) {
369     QualType InitType = Init->getType();
370     QualType DeclarationType = VDecl->getType();
371     if (!DeclarationType.isNull() && DeclarationType->isReferenceType())
372       DeclarationType = DeclarationType.getNonReferenceType();
373 
374     if (InitType.isNull() || DeclarationType.isNull() ||
375         !Context->hasSameUnqualifiedType(DeclarationType, InitType))
376       return false;
377   }
378 
379   switch (Init->getStmtClass()) {
380   case Stmt::ArraySubscriptExprClass: {
381     const auto *E = cast<ArraySubscriptExpr>(Init);
382     // We don't really care which array is used here. We check to make sure
383     // it was the correct one later, since the AST will traverse it next.
384     return isIndexInSubscriptExpr(E->getIdx(), IndexVar);
385   }
386 
387   case Stmt::UnaryOperatorClass:
388     return isDereferenceOfUop(cast<UnaryOperator>(Init), IndexVar);
389 
390   case Stmt::CXXOperatorCallExprClass: {
391     const auto *OpCall = cast<CXXOperatorCallExpr>(Init);
392     if (OpCall->getOperator() == OO_Star)
393       return isDereferenceOfOpCall(OpCall, IndexVar);
394     if (OpCall->getOperator() == OO_Subscript) {
395       assert(OpCall->getNumArgs() == 2);
396       return isIndexInSubscriptExpr(OpCall->getArg(1), IndexVar);
397     }
398     break;
399   }
400 
401   case Stmt::CXXMemberCallExprClass: {
402     const auto *MemCall = cast<CXXMemberCallExpr>(Init);
403     // This check is needed because getMethodDecl can return nullptr if the
404     // callee is a member function pointer.
405     const auto *MDecl = MemCall->getMethodDecl();
406     if (MDecl && !isa<CXXConversionDecl>(MDecl) &&
407         MDecl->getNameAsString() == "at" && MemCall->getNumArgs() == 1) {
408       return isIndexInSubscriptExpr(MemCall->getArg(0), IndexVar);
409     }
410     return false;
411   }
412 
413   default:
414     break;
415   }
416   return false;
417 }
418 
419 /// Determines whether the bound of a for loop condition expression is
420 /// the same as the statically computable size of ArrayType.
421 ///
422 /// Given
423 /// \code
424 ///   const int N = 5;
425 ///   int arr[N];
426 /// \endcode
427 /// This is intended to permit
428 /// \code
429 ///   for (int i = 0; i < N; ++i) {  /* use arr[i] */ }
430 ///   for (int i = 0; i < arraysize(arr); ++i) { /* use arr[i] */ }
431 /// \endcode
arrayMatchesBoundExpr(ASTContext * Context,const QualType & ArrayType,const Expr * ConditionExpr)432 static bool arrayMatchesBoundExpr(ASTContext *Context,
433                                   const QualType &ArrayType,
434                                   const Expr *ConditionExpr) {
435   if (!ConditionExpr || ConditionExpr->isValueDependent())
436     return false;
437   const ConstantArrayType *ConstType =
438       Context->getAsConstantArrayType(ArrayType);
439   if (!ConstType)
440     return false;
441   Optional<llvm::APSInt> ConditionSize =
442       ConditionExpr->getIntegerConstantExpr(*Context);
443   if (!ConditionSize)
444     return false;
445   llvm::APSInt ArraySize(ConstType->getSize());
446   return llvm::APSInt::isSameValue(*ConditionSize, ArraySize);
447 }
448 
ForLoopIndexUseVisitor(ASTContext * Context,const VarDecl * IndexVar,const VarDecl * EndVar,const Expr * ContainerExpr,const Expr * ArrayBoundExpr,bool ContainerNeedsDereference)449 ForLoopIndexUseVisitor::ForLoopIndexUseVisitor(ASTContext *Context,
450                                                const VarDecl *IndexVar,
451                                                const VarDecl *EndVar,
452                                                const Expr *ContainerExpr,
453                                                const Expr *ArrayBoundExpr,
454                                                bool ContainerNeedsDereference)
455     : Context(Context), IndexVar(IndexVar), EndVar(EndVar),
456       ContainerExpr(ContainerExpr), ArrayBoundExpr(ArrayBoundExpr),
457       ContainerNeedsDereference(ContainerNeedsDereference),
458       OnlyUsedAsIndex(true), AliasDecl(nullptr),
459       ConfidenceLevel(Confidence::CL_Safe), NextStmtParent(nullptr),
460       CurrStmtParent(nullptr), ReplaceWithAliasUse(false),
461       AliasFromForInit(false) {
462   if (ContainerExpr)
463     addComponent(ContainerExpr);
464 }
465 
findAndVerifyUsages(const Stmt * Body)466 bool ForLoopIndexUseVisitor::findAndVerifyUsages(const Stmt *Body) {
467   TraverseStmt(const_cast<Stmt *>(Body));
468   return OnlyUsedAsIndex && ContainerExpr;
469 }
470 
addComponents(const ComponentVector & Components)471 void ForLoopIndexUseVisitor::addComponents(const ComponentVector &Components) {
472   // FIXME: add sort(on ID)+unique to avoid extra work.
473   for (const auto &I : Components)
474     addComponent(I);
475 }
476 
addComponent(const Expr * E)477 void ForLoopIndexUseVisitor::addComponent(const Expr *E) {
478   llvm::FoldingSetNodeID ID;
479   const Expr *Node = E->IgnoreParenImpCasts();
480   Node->Profile(ID, *Context, true);
481   DependentExprs.push_back(std::make_pair(Node, ID));
482 }
483 
addUsage(const Usage & U)484 void ForLoopIndexUseVisitor::addUsage(const Usage &U) {
485   SourceLocation Begin = U.Range.getBegin();
486   if (Begin.isMacroID())
487     Begin = Context->getSourceManager().getSpellingLoc(Begin);
488 
489   if (UsageLocations.insert(Begin).second)
490     Usages.push_back(U);
491 }
492 
493 /// If the unary operator is a dereference of IndexVar, include it
494 /// as a valid usage and prune the traversal.
495 ///
496 /// For example, if container.begin() and container.end() both return pointers
497 /// to int, this makes sure that the initialization for `k` is not counted as an
498 /// unconvertible use of the iterator `i`.
499 /// \code
500 ///   for (int *i = container.begin(), *e = container.end(); i != e; ++i) {
501 ///     int k = *i + 2;
502 ///   }
503 /// \endcode
TraverseUnaryOperator(UnaryOperator * Uop)504 bool ForLoopIndexUseVisitor::TraverseUnaryOperator(UnaryOperator *Uop) {
505   // If we dereference an iterator that's actually a pointer, count the
506   // occurrence.
507   if (isDereferenceOfUop(Uop, IndexVar)) {
508     addUsage(Usage(Uop));
509     return true;
510   }
511 
512   return VisitorBase::TraverseUnaryOperator(Uop);
513 }
514 
515 /// If the member expression is operator-> (overloaded or not) on
516 /// IndexVar, include it as a valid usage and prune the traversal.
517 ///
518 /// For example, given
519 /// \code
520 ///   struct Foo { int bar(); int x; };
521 ///   vector<Foo> v;
522 /// \endcode
523 /// the following uses will be considered convertible:
524 /// \code
525 ///   for (vector<Foo>::iterator i = v.begin(), e = v.end(); i != e; ++i) {
526 ///     int b = i->bar();
527 ///     int k = i->x + 1;
528 ///   }
529 /// \endcode
530 /// though
531 /// \code
532 ///   for (vector<Foo>::iterator i = v.begin(), e = v.end(); i != e; ++i) {
533 ///     int k = i.insert(1);
534 ///   }
535 ///   for (vector<Foo>::iterator i = v.begin(), e = v.end(); i != e; ++i) {
536 ///     int b = e->bar();
537 ///   }
538 /// \endcode
539 /// will not.
TraverseMemberExpr(MemberExpr * Member)540 bool ForLoopIndexUseVisitor::TraverseMemberExpr(MemberExpr *Member) {
541   const Expr *Base = Member->getBase();
542   const DeclRefExpr *Obj = getDeclRef(Base);
543   const Expr *ResultExpr = Member;
544   QualType ExprType;
545   if (const auto *Call =
546           dyn_cast<CXXOperatorCallExpr>(Base->IgnoreParenImpCasts())) {
547     // If operator->() is a MemberExpr containing a CXXOperatorCallExpr, then
548     // the MemberExpr does not have the expression we want. We therefore catch
549     // that instance here.
550     // For example, if vector<Foo>::iterator defines operator->(), then the
551     // example `i->bar()` at the top of this function is a CXXMemberCallExpr
552     // referring to `i->` as the member function called. We want just `i`, so
553     // we take the argument to operator->() as the base object.
554     if (Call->getOperator() == OO_Arrow) {
555       assert(Call->getNumArgs() == 1 &&
556              "Operator-> takes more than one argument");
557       Obj = getDeclRef(Call->getArg(0));
558       ResultExpr = Obj;
559       ExprType = Call->getCallReturnType(*Context);
560     }
561   }
562 
563   if (Obj && exprReferencesVariable(IndexVar, Obj)) {
564     // Member calls on the iterator with '.' are not allowed.
565     if (!Member->isArrow()) {
566       OnlyUsedAsIndex = false;
567       return true;
568     }
569 
570     if (ExprType.isNull())
571       ExprType = Obj->getType();
572 
573     if (!ExprType->isPointerType())
574       return false;
575 
576     // FIXME: This works around not having the location of the arrow operator.
577     // Consider adding OperatorLoc to MemberExpr?
578     SourceLocation ArrowLoc = Lexer::getLocForEndOfToken(
579         Base->getExprLoc(), 0, Context->getSourceManager(),
580         Context->getLangOpts());
581     // If something complicated is happening (i.e. the next token isn't an
582     // arrow), give up on making this work.
583     if (ArrowLoc.isValid()) {
584       addUsage(Usage(ResultExpr, Usage::UK_MemberThroughArrow,
585                      SourceRange(Base->getExprLoc(), ArrowLoc)));
586       return true;
587     }
588   }
589   return VisitorBase::TraverseMemberExpr(Member);
590 }
591 
592 /// If a member function call is the at() accessor on the container with
593 /// IndexVar as the single argument, include it as a valid usage and prune
594 /// the traversal.
595 ///
596 /// Member calls on other objects will not be permitted.
597 /// Calls on the iterator object are not permitted, unless done through
598 /// operator->(). The one exception is allowing vector::at() for pseudoarrays.
TraverseCXXMemberCallExpr(CXXMemberCallExpr * MemberCall)599 bool ForLoopIndexUseVisitor::TraverseCXXMemberCallExpr(
600     CXXMemberCallExpr *MemberCall) {
601   auto *Member =
602       dyn_cast<MemberExpr>(MemberCall->getCallee()->IgnoreParenImpCasts());
603   if (!Member)
604     return VisitorBase::TraverseCXXMemberCallExpr(MemberCall);
605 
606   // We specifically allow an accessor named "at" to let STL in, though
607   // this is restricted to pseudo-arrays by requiring a single, integer
608   // argument.
609   const IdentifierInfo *Ident = Member->getMemberDecl()->getIdentifier();
610   if (Ident && Ident->isStr("at") && MemberCall->getNumArgs() == 1) {
611     if (isIndexInSubscriptExpr(Context, MemberCall->getArg(0), IndexVar,
612                                Member->getBase(), ContainerExpr,
613                                ContainerNeedsDereference)) {
614       addUsage(Usage(MemberCall));
615       return true;
616     }
617   }
618 
619   if (containsExpr(Context, &DependentExprs, Member->getBase()))
620     ConfidenceLevel.lowerTo(Confidence::CL_Risky);
621 
622   return VisitorBase::TraverseCXXMemberCallExpr(MemberCall);
623 }
624 
625 /// If an overloaded operator call is a dereference of IndexVar or
626 /// a subscript of the container with IndexVar as the single argument,
627 /// include it as a valid usage and prune the traversal.
628 ///
629 /// For example, given
630 /// \code
631 ///   struct Foo { int bar(); int x; };
632 ///   vector<Foo> v;
633 ///   void f(Foo);
634 /// \endcode
635 /// the following uses will be considered convertible:
636 /// \code
637 ///   for (vector<Foo>::iterator i = v.begin(), e = v.end(); i != e; ++i) {
638 ///     f(*i);
639 ///   }
640 ///   for (int i = 0; i < v.size(); ++i) {
641 ///      int i = v[i] + 1;
642 ///   }
643 /// \endcode
TraverseCXXOperatorCallExpr(CXXOperatorCallExpr * OpCall)644 bool ForLoopIndexUseVisitor::TraverseCXXOperatorCallExpr(
645     CXXOperatorCallExpr *OpCall) {
646   switch (OpCall->getOperator()) {
647   case OO_Star:
648     if (isDereferenceOfOpCall(OpCall, IndexVar)) {
649       addUsage(Usage(OpCall));
650       return true;
651     }
652     break;
653 
654   case OO_Subscript:
655     if (OpCall->getNumArgs() != 2)
656       break;
657     if (isIndexInSubscriptExpr(Context, OpCall->getArg(1), IndexVar,
658                                OpCall->getArg(0), ContainerExpr,
659                                ContainerNeedsDereference)) {
660       addUsage(Usage(OpCall));
661       return true;
662     }
663     break;
664 
665   default:
666     break;
667   }
668   return VisitorBase::TraverseCXXOperatorCallExpr(OpCall);
669 }
670 
671 /// If we encounter an array with IndexVar as the index of an
672 /// ArraySubscriptExpression, note it as a consistent usage and prune the
673 /// AST traversal.
674 ///
675 /// For example, given
676 /// \code
677 ///   const int N = 5;
678 ///   int arr[N];
679 /// \endcode
680 /// This is intended to permit
681 /// \code
682 ///   for (int i = 0; i < N; ++i) {  /* use arr[i] */ }
683 /// \endcode
684 /// but not
685 /// \code
686 ///   for (int i = 0; i < N; ++i) {  /* use notArr[i] */ }
687 /// \endcode
688 /// and further checking needs to be done later to ensure that exactly one array
689 /// is referenced.
TraverseArraySubscriptExpr(ArraySubscriptExpr * E)690 bool ForLoopIndexUseVisitor::TraverseArraySubscriptExpr(ArraySubscriptExpr *E) {
691   Expr *Arr = E->getBase();
692   if (!isIndexInSubscriptExpr(E->getIdx(), IndexVar))
693     return VisitorBase::TraverseArraySubscriptExpr(E);
694 
695   if ((ContainerExpr &&
696        !areSameExpr(Context, Arr->IgnoreParenImpCasts(),
697                     ContainerExpr->IgnoreParenImpCasts())) ||
698       !arrayMatchesBoundExpr(Context, Arr->IgnoreImpCasts()->getType(),
699                              ArrayBoundExpr)) {
700     // If we have already discovered the array being indexed and this isn't it
701     // or this array doesn't match, mark this loop as unconvertible.
702     OnlyUsedAsIndex = false;
703     return VisitorBase::TraverseArraySubscriptExpr(E);
704   }
705 
706   if (!ContainerExpr)
707     ContainerExpr = Arr;
708 
709   addUsage(Usage(E));
710   return true;
711 }
712 
713 /// If we encounter a reference to IndexVar in an unpruned branch of the
714 /// traversal, mark this loop as unconvertible.
715 ///
716 /// This determines the set of convertible loops: any usages of IndexVar
717 /// not explicitly considered convertible by this traversal will be caught by
718 /// this function.
719 ///
720 /// Additionally, if the container expression is more complex than just a
721 /// DeclRefExpr, and some part of it is appears elsewhere in the loop, lower
722 /// our confidence in the transformation.
723 ///
724 /// For example, these are not permitted:
725 /// \code
726 ///   for (int i = 0; i < N; ++i) {  printf("arr[%d] = %d", i, arr[i]); }
727 ///   for (vector<int>::iterator i = container.begin(), e = container.end();
728 ///        i != e; ++i)
729 ///     i.insert(0);
730 ///   for (vector<int>::iterator i = container.begin(), e = container.end();
731 ///        i != e; ++i)
732 ///     if (i + 1 != e)
733 ///       printf("%d", *i);
734 /// \endcode
735 ///
736 /// And these will raise the risk level:
737 /// \code
738 ///    int arr[10][20];
739 ///    int l = 5;
740 ///    for (int j = 0; j < 20; ++j)
741 ///      int k = arr[l][j] + l; // using l outside arr[l] is considered risky
742 ///    for (int i = 0; i < obj.getVector().size(); ++i)
743 ///      obj.foo(10); // using `obj` is considered risky
744 /// \endcode
VisitDeclRefExpr(DeclRefExpr * E)745 bool ForLoopIndexUseVisitor::VisitDeclRefExpr(DeclRefExpr *E) {
746   const ValueDecl *TheDecl = E->getDecl();
747   if (areSameVariable(IndexVar, TheDecl) ||
748       exprReferencesVariable(IndexVar, E) || areSameVariable(EndVar, TheDecl) ||
749       exprReferencesVariable(EndVar, E))
750     OnlyUsedAsIndex = false;
751   if (containsExpr(Context, &DependentExprs, E))
752     ConfidenceLevel.lowerTo(Confidence::CL_Risky);
753   return true;
754 }
755 
756 /// If the loop index is captured by a lambda, replace this capture
757 /// by the range-for loop variable.
758 ///
759 /// For example:
760 /// \code
761 ///   for (int i = 0; i < N; ++i) {
762 ///     auto f = [v, i](int k) {
763 ///       printf("%d\n", v[i] + k);
764 ///     };
765 ///     f(v[i]);
766 ///   }
767 /// \endcode
768 ///
769 /// Will be replaced by:
770 /// \code
771 ///   for (auto & elem : v) {
772 ///     auto f = [v, elem](int k) {
773 ///       printf("%d\n", elem + k);
774 ///     };
775 ///     f(elem);
776 ///   }
777 /// \endcode
TraverseLambdaCapture(LambdaExpr * LE,const LambdaCapture * C,Expr * Init)778 bool ForLoopIndexUseVisitor::TraverseLambdaCapture(LambdaExpr *LE,
779                                                    const LambdaCapture *C,
780                                                    Expr *Init) {
781   if (C->capturesVariable()) {
782     const VarDecl *VDecl = C->getCapturedVar();
783     if (areSameVariable(IndexVar, cast<ValueDecl>(VDecl))) {
784       // FIXME: if the index is captured, it will count as an usage and the
785       // alias (if any) won't work, because it is only used in case of having
786       // exactly one usage.
787       addUsage(Usage(nullptr,
788                      C->getCaptureKind() == LCK_ByCopy ? Usage::UK_CaptureByCopy
789                                                        : Usage::UK_CaptureByRef,
790                      C->getLocation()));
791     }
792   }
793   return VisitorBase::TraverseLambdaCapture(LE, C, Init);
794 }
795 
796 /// If we find that another variable is created just to refer to the loop
797 /// element, note it for reuse as the loop variable.
798 ///
799 /// See the comments for isAliasDecl.
VisitDeclStmt(DeclStmt * S)800 bool ForLoopIndexUseVisitor::VisitDeclStmt(DeclStmt *S) {
801   if (!AliasDecl && S->isSingleDecl() &&
802       isAliasDecl(Context, S->getSingleDecl(), IndexVar)) {
803     AliasDecl = S;
804     if (CurrStmtParent) {
805       if (isa<IfStmt>(CurrStmtParent) || isa<WhileStmt>(CurrStmtParent) ||
806           isa<SwitchStmt>(CurrStmtParent))
807         ReplaceWithAliasUse = true;
808       else if (isa<ForStmt>(CurrStmtParent)) {
809         if (cast<ForStmt>(CurrStmtParent)->getConditionVariableDeclStmt() == S)
810           ReplaceWithAliasUse = true;
811         else
812           // It's assumed S came the for loop's init clause.
813           AliasFromForInit = true;
814       }
815     }
816   }
817 
818   return true;
819 }
820 
TraverseStmt(Stmt * S)821 bool ForLoopIndexUseVisitor::TraverseStmt(Stmt *S) {
822   // If this is an initialization expression for a lambda capture, prune the
823   // traversal so that we don't end up diagnosing the contained DeclRefExpr as
824   // inconsistent usage. No need to record the usage here -- this is done in
825   // TraverseLambdaCapture().
826   if (const auto *LE = dyn_cast_or_null<LambdaExpr>(NextStmtParent)) {
827     // Any child of a LambdaExpr that isn't the body is an initialization
828     // expression.
829     if (S != LE->getBody()) {
830       return true;
831     }
832   }
833 
834   // All this pointer swapping is a mechanism for tracking immediate parentage
835   // of Stmts.
836   const Stmt *OldNextParent = NextStmtParent;
837   CurrStmtParent = NextStmtParent;
838   NextStmtParent = S;
839   bool Result = VisitorBase::TraverseStmt(S);
840   NextStmtParent = OldNextParent;
841   return Result;
842 }
843 
createIndexName()844 std::string VariableNamer::createIndexName() {
845   // FIXME: Add in naming conventions to handle:
846   //  - How to handle conflicts.
847   //  - An interactive process for naming.
848   std::string IteratorName;
849   StringRef ContainerName;
850   if (TheContainer)
851     ContainerName = TheContainer->getName();
852 
853   size_t Len = ContainerName.size();
854   if (Len > 1 && ContainerName.endswith(Style == NS_UpperCase ? "S" : "s")) {
855     IteratorName = std::string(ContainerName.substr(0, Len - 1));
856     // E.g.: (auto thing : things)
857     if (!declarationExists(IteratorName) || IteratorName == OldIndex->getName())
858       return IteratorName;
859   }
860 
861   if (Len > 2 && ContainerName.endswith(Style == NS_UpperCase ? "S_" : "s_")) {
862     IteratorName = std::string(ContainerName.substr(0, Len - 2));
863     // E.g.: (auto thing : things_)
864     if (!declarationExists(IteratorName) || IteratorName == OldIndex->getName())
865       return IteratorName;
866   }
867 
868   return std::string(OldIndex->getName());
869 }
870 
871 /// Determines whether or not the name \a Symbol conflicts with
872 /// language keywords or defined macros. Also checks if the name exists in
873 /// LoopContext, any of its parent contexts, or any of its child statements.
874 ///
875 /// We also check to see if the same identifier was generated by this loop
876 /// converter in a loop nested within SourceStmt.
declarationExists(StringRef Symbol)877 bool VariableNamer::declarationExists(StringRef Symbol) {
878   assert(Context != nullptr && "Expected an ASTContext");
879   IdentifierInfo &Ident = Context->Idents.get(Symbol);
880 
881   // Check if the symbol is not an identifier (ie. is a keyword or alias).
882   if (!isAnyIdentifier(Ident.getTokenID()))
883     return true;
884 
885   // Check for conflicting macro definitions.
886   if (Ident.hasMacroDefinition())
887     return true;
888 
889   // Determine if the symbol was generated in a parent context.
890   for (const Stmt *S = SourceStmt; S != nullptr; S = ReverseAST->lookup(S)) {
891     StmtGeneratedVarNameMap::const_iterator I = GeneratedDecls->find(S);
892     if (I != GeneratedDecls->end() && I->second == Symbol)
893       return true;
894   }
895 
896   // FIXME: Rather than detecting conflicts at their usages, we should check the
897   // parent context.
898   // For some reason, lookup() always returns the pair (NULL, NULL) because its
899   // StoredDeclsMap is not initialized (i.e. LookupPtr.getInt() is false inside
900   // of DeclContext::lookup()). Why is this?
901 
902   // Finally, determine if the symbol was used in the loop or a child context.
903   DeclFinderASTVisitor DeclFinder(std::string(Symbol), GeneratedDecls);
904   return DeclFinder.findUsages(SourceStmt);
905 }
906 
907 } // namespace modernize
908 } // namespace tidy
909 } // namespace clang
910