clang-tools  6.0.0
LoopConvertCheck.cpp
Go to the documentation of this file.
1 //===--- LoopConvertCheck.cpp - clang-tidy---------------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "LoopConvertCheck.h"
11 #include "clang/AST/ASTContext.h"
12 #include "clang/ASTMatchers/ASTMatchFinder.h"
13 #include "clang/Basic/LLVM.h"
14 #include "clang/Basic/LangOptions.h"
15 #include "clang/Basic/SourceLocation.h"
16 #include "clang/Basic/SourceManager.h"
17 #include "clang/Lex/Lexer.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/ADT/StringSwitch.h"
21 #include "llvm/Support/Casting.h"
22 #include <cassert>
23 #include <cstring>
24 #include <utility>
25 
26 using namespace clang::ast_matchers;
27 using namespace llvm;
28 
29 namespace clang {
30 namespace tidy {
31 namespace modernize {
32 
33 static const char LoopNameArray[] = "forLoopArray";
34 static const char LoopNameIterator[] = "forLoopIterator";
35 static const char LoopNamePseudoArray[] = "forLoopPseudoArray";
36 static const char ConditionBoundName[] = "conditionBound";
37 static const char ConditionVarName[] = "conditionVar";
38 static const char IncrementVarName[] = "incrementVar";
39 static const char InitVarName[] = "initVar";
40 static const char BeginCallName[] = "beginCall";
41 static const char EndCallName[] = "endCall";
42 static const char ConditionEndVarName[] = "conditionEndVar";
43 static const char EndVarName[] = "endVar";
44 static const char DerefByValueResultName[] = "derefByValueResult";
45 static const char DerefByRefResultName[] = "derefByRefResult";
46 
47 // shared matchers
48 static const TypeMatcher AnyType = anything();
49 
50 static const StatementMatcher IntegerComparisonMatcher =
51  expr(ignoringParenImpCasts(
52  declRefExpr(to(varDecl(hasType(isInteger())).bind(ConditionVarName)))));
53 
54 static const DeclarationMatcher InitToZeroMatcher =
55  varDecl(hasInitializer(ignoringParenImpCasts(integerLiteral(equals(0)))))
56  .bind(InitVarName);
57 
58 static const StatementMatcher IncrementVarMatcher =
59  declRefExpr(to(varDecl(hasType(isInteger())).bind(IncrementVarName)));
60 
61 /// \brief The matcher for loops over arrays.
62 ///
63 /// In this general example, assuming 'j' and 'k' are of integral type:
64 /// \code
65 /// for (int i = 0; j < 3 + 2; ++k) { ... }
66 /// \endcode
67 /// The following string identifiers are bound to these parts of the AST:
68 /// ConditionVarName: 'j' (as a VarDecl)
69 /// ConditionBoundName: '3 + 2' (as an Expr)
70 /// InitVarName: 'i' (as a VarDecl)
71 /// IncrementVarName: 'k' (as a VarDecl)
72 /// LoopName: The entire for loop (as a ForStmt)
73 ///
74 /// Client code will need to make sure that:
75 /// - The three index variables identified by the matcher are the same
76 /// VarDecl.
77 /// - The index variable is only used as an array index.
78 /// - All arrays indexed by the loop are the same.
79 StatementMatcher makeArrayLoopMatcher() {
80  StatementMatcher ArrayBoundMatcher =
81  expr(hasType(isInteger())).bind(ConditionBoundName);
82 
83  return forStmt(
84  unless(isInTemplateInstantiation()),
85  hasLoopInit(declStmt(hasSingleDecl(InitToZeroMatcher))),
86  hasCondition(anyOf(
87  binaryOperator(hasOperatorName("<"),
88  hasLHS(IntegerComparisonMatcher),
89  hasRHS(ArrayBoundMatcher)),
90  binaryOperator(hasOperatorName(">"), hasLHS(ArrayBoundMatcher),
91  hasRHS(IntegerComparisonMatcher)))),
92  hasIncrement(unaryOperator(hasOperatorName("++"),
93  hasUnaryOperand(IncrementVarMatcher))))
94  .bind(LoopNameArray);
95 }
96 
97 /// \brief The matcher used for iterator-based for loops.
98 ///
99 /// This matcher is more flexible than array-based loops. It will match
100 /// catch loops of the following textual forms (regardless of whether the
101 /// iterator type is actually a pointer type or a class type):
102 ///
103 /// Assuming f, g, and h are of type containerType::iterator,
104 /// \code
105 /// for (containerType::iterator it = container.begin(),
106 /// e = createIterator(); f != g; ++h) { ... }
107 /// for (containerType::iterator it = container.begin();
108 /// f != anotherContainer.end(); ++h) { ... }
109 /// \endcode
110 /// The following string identifiers are bound to the parts of the AST:
111 /// InitVarName: 'it' (as a VarDecl)
112 /// ConditionVarName: 'f' (as a VarDecl)
113 /// LoopName: The entire for loop (as a ForStmt)
114 /// In the first example only:
115 /// EndVarName: 'e' (as a VarDecl)
116 /// ConditionEndVarName: 'g' (as a VarDecl)
117 /// In the second example only:
118 /// EndCallName: 'container.end()' (as a CXXMemberCallExpr)
119 ///
120 /// Client code will need to make sure that:
121 /// - The iterator variables 'it', 'f', and 'h' are the same.
122 /// - The two containers on which 'begin' and 'end' are called are the same.
123 /// - If the end iterator variable 'g' is defined, it is the same as 'f'.
124 StatementMatcher makeIteratorLoopMatcher() {
125  StatementMatcher BeginCallMatcher =
126  cxxMemberCallExpr(
127  argumentCountIs(0),
128  callee(cxxMethodDecl(anyOf(hasName("begin"), hasName("cbegin")))))
129  .bind(BeginCallName);
130 
131  DeclarationMatcher InitDeclMatcher =
132  varDecl(hasInitializer(anyOf(ignoringParenImpCasts(BeginCallMatcher),
133  materializeTemporaryExpr(
134  ignoringParenImpCasts(BeginCallMatcher)),
135  hasDescendant(BeginCallMatcher))))
136  .bind(InitVarName);
137 
138  DeclarationMatcher EndDeclMatcher =
139  varDecl(hasInitializer(anything())).bind(EndVarName);
140 
141  StatementMatcher EndCallMatcher = cxxMemberCallExpr(
142  argumentCountIs(0),
143  callee(cxxMethodDecl(anyOf(hasName("end"), hasName("cend")))));
144 
145  StatementMatcher IteratorBoundMatcher =
146  expr(anyOf(ignoringParenImpCasts(
147  declRefExpr(to(varDecl().bind(ConditionEndVarName)))),
148  ignoringParenImpCasts(expr(EndCallMatcher).bind(EndCallName)),
149  materializeTemporaryExpr(ignoringParenImpCasts(
150  expr(EndCallMatcher).bind(EndCallName)))));
151 
152  StatementMatcher IteratorComparisonMatcher = expr(
153  ignoringParenImpCasts(declRefExpr(to(varDecl().bind(ConditionVarName)))));
154 
155  auto OverloadedNEQMatcher = ignoringImplicit(
156  cxxOperatorCallExpr(hasOverloadedOperatorName("!="), argumentCountIs(2),
157  hasArgument(0, IteratorComparisonMatcher),
158  hasArgument(1, IteratorBoundMatcher)));
159 
160  // This matcher tests that a declaration is a CXXRecordDecl that has an
161  // overloaded operator*(). If the operator*() returns by value instead of by
162  // reference then the return type is tagged with DerefByValueResultName.
163  internal::Matcher<VarDecl> TestDerefReturnsByValue =
164  hasType(hasUnqualifiedDesugaredType(
165  recordType(hasDeclaration(cxxRecordDecl(hasMethod(allOf(
166  hasOverloadedOperatorName("*"),
167  anyOf(
168  // Tag the return type if it's by value.
169  returns(qualType(unless(hasCanonicalType(referenceType())))
170  .bind(DerefByValueResultName)),
171  returns(
172  // Skip loops where the iterator's operator* returns an
173  // rvalue reference. This is just weird.
174  qualType(unless(hasCanonicalType(rValueReferenceType())))
175  .bind(DerefByRefResultName))))))))));
176 
177  return forStmt(
178  unless(isInTemplateInstantiation()),
179  hasLoopInit(anyOf(declStmt(declCountIs(2),
180  containsDeclaration(0, InitDeclMatcher),
181  containsDeclaration(1, EndDeclMatcher)),
182  declStmt(hasSingleDecl(InitDeclMatcher)))),
183  hasCondition(
184  anyOf(binaryOperator(hasOperatorName("!="),
185  hasLHS(IteratorComparisonMatcher),
186  hasRHS(IteratorBoundMatcher)),
187  binaryOperator(hasOperatorName("!="),
188  hasLHS(IteratorBoundMatcher),
189  hasRHS(IteratorComparisonMatcher)),
190  OverloadedNEQMatcher)),
191  hasIncrement(anyOf(
192  unaryOperator(hasOperatorName("++"),
193  hasUnaryOperand(declRefExpr(
194  to(varDecl(hasType(pointsTo(AnyType)))
195  .bind(IncrementVarName))))),
196  cxxOperatorCallExpr(
197  hasOverloadedOperatorName("++"),
198  hasArgument(
199  0, declRefExpr(to(varDecl(TestDerefReturnsByValue)
200  .bind(IncrementVarName))))))))
201  .bind(LoopNameIterator);
202 }
203 
204 /// \brief The matcher used for array-like containers (pseudoarrays).
205 ///
206 /// This matcher is more flexible than array-based loops. It will match
207 /// loops of the following textual forms (regardless of whether the
208 /// iterator type is actually a pointer type or a class type):
209 ///
210 /// Assuming f, g, and h are of type containerType::iterator,
211 /// \code
212 /// for (int i = 0, j = container.size(); f < g; ++h) { ... }
213 /// for (int i = 0; f < container.size(); ++h) { ... }
214 /// \endcode
215 /// The following string identifiers are bound to the parts of the AST:
216 /// InitVarName: 'i' (as a VarDecl)
217 /// ConditionVarName: 'f' (as a VarDecl)
218 /// LoopName: The entire for loop (as a ForStmt)
219 /// In the first example only:
220 /// EndVarName: 'j' (as a VarDecl)
221 /// ConditionEndVarName: 'g' (as a VarDecl)
222 /// In the second example only:
223 /// EndCallName: 'container.size()' (as a CXXMemberCallExpr)
224 ///
225 /// Client code will need to make sure that:
226 /// - The index variables 'i', 'f', and 'h' are the same.
227 /// - The containers on which 'size()' is called is the container indexed.
228 /// - The index variable is only used in overloaded operator[] or
229 /// container.at().
230 /// - If the end iterator variable 'g' is defined, it is the same as 'j'.
231 /// - The container's iterators would not be invalidated during the loop.
232 StatementMatcher makePseudoArrayLoopMatcher() {
233  // Test that the incoming type has a record declaration that has methods
234  // called 'begin' and 'end'. If the incoming type is const, then make sure
235  // these methods are also marked const.
236  //
237  // FIXME: To be completely thorough this matcher should also ensure the
238  // return type of begin/end is an iterator that dereferences to the same as
239  // what operator[] or at() returns. Such a test isn't likely to fail except
240  // for pathological cases.
241  //
242  // FIXME: Also, a record doesn't necessarily need begin() and end(). Free
243  // functions called begin() and end() taking the container as an argument
244  // are also allowed.
245  TypeMatcher RecordWithBeginEnd = qualType(anyOf(
246  qualType(
247  isConstQualified(),
248  hasUnqualifiedDesugaredType(recordType(hasDeclaration(cxxRecordDecl(
249  hasMethod(cxxMethodDecl(hasName("begin"), isConst())),
250  hasMethod(cxxMethodDecl(hasName("end"),
251  isConst())))) // hasDeclaration
252  ))), // qualType
253  qualType(unless(isConstQualified()),
254  hasUnqualifiedDesugaredType(recordType(hasDeclaration(
255  cxxRecordDecl(hasMethod(hasName("begin")),
256  hasMethod(hasName("end"))))))) // qualType
257  ));
258 
259  StatementMatcher SizeCallMatcher = cxxMemberCallExpr(
260  argumentCountIs(0),
261  callee(cxxMethodDecl(anyOf(hasName("size"), hasName("length")))),
262  on(anyOf(hasType(pointsTo(RecordWithBeginEnd)),
263  hasType(RecordWithBeginEnd))));
264 
265  StatementMatcher EndInitMatcher =
266  expr(anyOf(ignoringParenImpCasts(expr(SizeCallMatcher).bind(EndCallName)),
267  explicitCastExpr(hasSourceExpression(ignoringParenImpCasts(
268  expr(SizeCallMatcher).bind(EndCallName))))));
269 
270  DeclarationMatcher EndDeclMatcher =
271  varDecl(hasInitializer(EndInitMatcher)).bind(EndVarName);
272 
273  StatementMatcher IndexBoundMatcher =
274  expr(anyOf(ignoringParenImpCasts(declRefExpr(to(
275  varDecl(hasType(isInteger())).bind(ConditionEndVarName)))),
276  EndInitMatcher));
277 
278  return forStmt(
279  unless(isInTemplateInstantiation()),
280  hasLoopInit(
281  anyOf(declStmt(declCountIs(2),
282  containsDeclaration(0, InitToZeroMatcher),
283  containsDeclaration(1, EndDeclMatcher)),
284  declStmt(hasSingleDecl(InitToZeroMatcher)))),
285  hasCondition(anyOf(
286  binaryOperator(hasOperatorName("<"),
287  hasLHS(IntegerComparisonMatcher),
288  hasRHS(IndexBoundMatcher)),
289  binaryOperator(hasOperatorName(">"), hasLHS(IndexBoundMatcher),
290  hasRHS(IntegerComparisonMatcher)))),
291  hasIncrement(unaryOperator(hasOperatorName("++"),
292  hasUnaryOperand(IncrementVarMatcher))))
293  .bind(LoopNamePseudoArray);
294 }
295 
296 /// \brief Determine whether Init appears to be an initializing an iterator.
297 ///
298 /// If it is, returns the object whose begin() or end() method is called, and
299 /// the output parameter isArrow is set to indicate whether the initialization
300 /// is called via . or ->.
301 static const Expr *getContainerFromBeginEndCall(const Expr *Init, bool IsBegin,
302  bool *IsArrow) {
303  // FIXME: Maybe allow declaration/initialization outside of the for loop.
304  const auto *TheCall =
305  dyn_cast_or_null<CXXMemberCallExpr>(digThroughConstructors(Init));
306  if (!TheCall || TheCall->getNumArgs() != 0)
307  return nullptr;
308 
309  const auto *Member = dyn_cast<MemberExpr>(TheCall->getCallee());
310  if (!Member)
311  return nullptr;
312  StringRef Name = Member->getMemberDecl()->getName();
313  StringRef TargetName = IsBegin ? "begin" : "end";
314  StringRef ConstTargetName = IsBegin ? "cbegin" : "cend";
315  if (Name != TargetName && Name != ConstTargetName)
316  return nullptr;
317 
318  const Expr *SourceExpr = Member->getBase();
319  if (!SourceExpr)
320  return nullptr;
321 
322  *IsArrow = Member->isArrow();
323  return SourceExpr;
324 }
325 
326 /// \brief Determines the container whose begin() and end() functions are called
327 /// for an iterator-based loop.
328 ///
329 /// BeginExpr must be a member call to a function named "begin()", and EndExpr
330 /// must be a member.
331 static const Expr *findContainer(ASTContext *Context, const Expr *BeginExpr,
332  const Expr *EndExpr,
333  bool *ContainerNeedsDereference) {
334  // Now that we know the loop variable and test expression, make sure they are
335  // valid.
336  bool BeginIsArrow = false;
337  bool EndIsArrow = false;
338  const Expr *BeginContainerExpr =
339  getContainerFromBeginEndCall(BeginExpr, /*IsBegin=*/true, &BeginIsArrow);
340  if (!BeginContainerExpr)
341  return nullptr;
342 
343  const Expr *EndContainerExpr =
344  getContainerFromBeginEndCall(EndExpr, /*IsBegin=*/false, &EndIsArrow);
345  // Disallow loops that try evil things like this (note the dot and arrow):
346  // for (IteratorType It = Obj.begin(), E = Obj->end(); It != E; ++It) { }
347  if (!EndContainerExpr || BeginIsArrow != EndIsArrow ||
348  !areSameExpr(Context, EndContainerExpr, BeginContainerExpr))
349  return nullptr;
350 
351  *ContainerNeedsDereference = BeginIsArrow;
352  return BeginContainerExpr;
353 }
354 
355 /// \brief Obtain the original source code text from a SourceRange.
356 static StringRef getStringFromRange(SourceManager &SourceMgr,
357  const LangOptions &LangOpts,
358  SourceRange Range) {
359  if (SourceMgr.getFileID(Range.getBegin()) !=
360  SourceMgr.getFileID(Range.getEnd())) {
361  return StringRef(); // Empty string.
362  }
363 
364  return Lexer::getSourceText(CharSourceRange(Range, true), SourceMgr,
365  LangOpts);
366 }
367 
368 /// \brief If the given expression is actually a DeclRefExpr or a MemberExpr,
369 /// find and return the underlying ValueDecl; otherwise, return NULL.
370 static const ValueDecl *getReferencedVariable(const Expr *E) {
371  if (const DeclRefExpr *DRE = getDeclRef(E))
372  return dyn_cast<VarDecl>(DRE->getDecl());
373  if (const auto *Mem = dyn_cast<MemberExpr>(E->IgnoreParenImpCasts()))
374  return dyn_cast<FieldDecl>(Mem->getMemberDecl());
375  return nullptr;
376 }
377 
378 /// \brief Returns true when the given expression is a member expression
379 /// whose base is `this` (implicitly or not).
380 static bool isDirectMemberExpr(const Expr *E) {
381  if (const auto *Member = dyn_cast<MemberExpr>(E->IgnoreParenImpCasts()))
382  return isa<CXXThisExpr>(Member->getBase()->IgnoreParenImpCasts());
383  return false;
384 }
385 
386 /// \brief Given an expression that represents an usage of an element from the
387 /// containter that we are iterating over, returns false when it can be
388 /// guaranteed this element cannot be modified as a result of this usage.
389 static bool canBeModified(ASTContext *Context, const Expr *E) {
390  if (E->getType().isConstQualified())
391  return false;
392  auto Parents = Context->getParents(*E);
393  if (Parents.size() != 1)
394  return true;
395  if (const auto *Cast = Parents[0].get<ImplicitCastExpr>()) {
396  if ((Cast->getCastKind() == CK_NoOp &&
397  Cast->getType() == E->getType().withConst()) ||
398  (Cast->getCastKind() == CK_LValueToRValue &&
399  !Cast->getType().isNull() && Cast->getType()->isFundamentalType()))
400  return false;
401  }
402  // FIXME: Make this function more generic.
403  return true;
404 }
405 
406 /// \brief Returns true when it can be guaranteed that the elements of the
407 /// container are not being modified.
408 static bool usagesAreConst(ASTContext *Context, const UsageResult &Usages) {
409  for (const Usage &U : Usages) {
410  // Lambda captures are just redeclarations (VarDecl) of the same variable,
411  // not expressions. If we want to know if a variable that is captured by
412  // reference can be modified in an usage inside the lambda's body, we need
413  // to find the expression corresponding to that particular usage, later in
414  // this loop.
415  if (U.Kind != Usage::UK_CaptureByCopy && U.Kind != Usage::UK_CaptureByRef &&
416  canBeModified(Context, U.Expression))
417  return false;
418  }
419  return true;
420 }
421 
422 /// \brief Returns true if the elements of the container are never accessed
423 /// by reference.
424 static bool usagesReturnRValues(const UsageResult &Usages) {
425  for (const auto &U : Usages) {
426  if (U.Expression && !U.Expression->isRValue())
427  return false;
428  }
429  return true;
430 }
431 
432 /// \brief Returns true if the container is const-qualified.
433 static bool containerIsConst(const Expr *ContainerExpr, bool Dereference) {
434  if (const auto *VDec = getReferencedVariable(ContainerExpr)) {
435  QualType CType = VDec->getType();
436  if (Dereference) {
437  if (!CType->isPointerType())
438  return false;
439  CType = CType->getPointeeType();
440  }
441  // If VDec is a reference to a container, Dereference is false,
442  // but we still need to check the const-ness of the underlying container
443  // type.
444  CType = CType.getNonReferenceType();
445  return CType.isConstQualified();
446  }
447  return false;
448 }
449 
450 LoopConvertCheck::RangeDescriptor::RangeDescriptor()
451  : ContainerNeedsDereference(false), DerefByConstRef(false),
452  DerefByValue(false) {}
453 
454 LoopConvertCheck::LoopConvertCheck(StringRef Name, ClangTidyContext *Context)
455  : ClangTidyCheck(Name, Context), TUInfo(new TUTrackingInfo),
456  MaxCopySize(std::stoull(Options.get("MaxCopySize", "16"))),
457  MinConfidence(StringSwitch<Confidence::Level>(
458  Options.get("MinConfidence", "reasonable"))
459  .Case("safe", Confidence::CL_Safe)
460  .Case("risky", Confidence::CL_Risky)
461  .Default(Confidence::CL_Reasonable)),
462  NamingStyle(StringSwitch<VariableNamer::NamingStyle>(
463  Options.get("NamingStyle", "CamelCase"))
464  .Case("camelBack", VariableNamer::NS_CamelBack)
465  .Case("lower_case", VariableNamer::NS_LowerCase)
466  .Case("UPPER_CASE", VariableNamer::NS_UpperCase)
467  .Default(VariableNamer::NS_CamelCase)) {}
468 
470  Options.store(Opts, "MaxCopySize", std::to_string(MaxCopySize));
471  SmallVector<std::string, 3> Confs{"risky", "reasonable", "safe"};
472  Options.store(Opts, "MinConfidence", Confs[static_cast<int>(MinConfidence)]);
473 
474  SmallVector<std::string, 4> Styles{"camelBack", "CamelCase", "lower_case",
475  "UPPER_CASE"};
476  Options.store(Opts, "NamingStyle", Styles[static_cast<int>(NamingStyle)]);
477 }
478 
479 void LoopConvertCheck::registerMatchers(MatchFinder *Finder) {
480  // Only register the matchers for C++. Because this checker is used for
481  // modernization, it is reasonable to run it on any C++ standard with the
482  // assumption the user is trying to modernize their codebase.
483  if (!getLangOpts().CPlusPlus)
484  return;
485 
486  Finder->addMatcher(makeArrayLoopMatcher(), this);
487  Finder->addMatcher(makeIteratorLoopMatcher(), this);
488  Finder->addMatcher(makePseudoArrayLoopMatcher(), this);
489 }
490 
491 /// \brief Given the range of a single declaration, such as:
492 /// \code
493 /// unsigned &ThisIsADeclarationThatCanSpanSeveralLinesOfCode =
494 /// InitializationValues[I];
495 /// next_instruction;
496 /// \endcode
497 /// Finds the range that has to be erased to remove this declaration without
498 /// leaving empty lines, by extending the range until the beginning of the
499 /// next instruction.
500 ///
501 /// We need to delete a potential newline after the deleted alias, as
502 /// clang-format will leave empty lines untouched. For all other formatting we
503 /// rely on clang-format to fix it.
504 void LoopConvertCheck::getAliasRange(SourceManager &SM, SourceRange &Range) {
505  bool Invalid = false;
506  const char *TextAfter =
507  SM.getCharacterData(Range.getEnd().getLocWithOffset(1), &Invalid);
508  if (Invalid)
509  return;
510  unsigned Offset = std::strspn(TextAfter, " \t\r\n");
511  Range =
512  SourceRange(Range.getBegin(), Range.getEnd().getLocWithOffset(Offset));
513 }
514 
515 /// \brief Computes the changes needed to convert a given for loop, and
516 /// applies them.
517 void LoopConvertCheck::doConversion(
518  ASTContext *Context, const VarDecl *IndexVar,
519  const ValueDecl *MaybeContainer, const UsageResult &Usages,
520  const DeclStmt *AliasDecl, bool AliasUseRequired, bool AliasFromForInit,
521  const ForStmt *Loop, RangeDescriptor Descriptor) {
522  auto Diag = diag(Loop->getForLoc(), "use range-based for loop instead");
523 
524  std::string VarName;
525  bool VarNameFromAlias = (Usages.size() == 1) && AliasDecl;
526  bool AliasVarIsRef = false;
527  bool CanCopy = true;
528 
529  if (VarNameFromAlias) {
530  const auto *AliasVar = cast<VarDecl>(AliasDecl->getSingleDecl());
531  VarName = AliasVar->getName().str();
532 
533  // Use the type of the alias if it's not the same
534  QualType AliasVarType = AliasVar->getType();
535  assert(!AliasVarType.isNull() && "Type in VarDecl is null");
536  if (AliasVarType->isReferenceType()) {
537  AliasVarType = AliasVarType.getNonReferenceType();
538  AliasVarIsRef = true;
539  }
540  if (Descriptor.ElemType.isNull() ||
541  !Context->hasSameUnqualifiedType(AliasVarType, Descriptor.ElemType))
542  Descriptor.ElemType = AliasVarType;
543 
544  // We keep along the entire DeclStmt to keep the correct range here.
545  SourceRange ReplaceRange = AliasDecl->getSourceRange();
546 
547  std::string ReplacementText;
548  if (AliasUseRequired) {
549  ReplacementText = VarName;
550  } else if (AliasFromForInit) {
551  // FIXME: Clang includes the location of the ';' but only for DeclStmt's
552  // in a for loop's init clause. Need to put this ';' back while removing
553  // the declaration of the alias variable. This is probably a bug.
554  ReplacementText = ";";
555  } else {
556  // Avoid leaving empty lines or trailing whitespaces.
557  getAliasRange(Context->getSourceManager(), ReplaceRange);
558  }
559 
560  Diag << FixItHint::CreateReplacement(
561  CharSourceRange::getTokenRange(ReplaceRange), ReplacementText);
562  // No further replacements are made to the loop, since the iterator or index
563  // was used exactly once - in the initialization of AliasVar.
564  } else {
565  VariableNamer Namer(&TUInfo->getGeneratedDecls(),
566  &TUInfo->getParentFinder().getStmtToParentStmtMap(),
567  Loop, IndexVar, MaybeContainer, Context, NamingStyle);
568  VarName = Namer.createIndexName();
569  // First, replace all usages of the array subscript expression with our new
570  // variable.
571  for (const auto &Usage : Usages) {
572  std::string ReplaceText;
573  SourceRange Range = Usage.Range;
574  if (Usage.Expression) {
575  // If this is an access to a member through the arrow operator, after
576  // the replacement it must be accessed through the '.' operator.
577  ReplaceText = Usage.Kind == Usage::UK_MemberThroughArrow ? VarName + "."
578  : VarName;
579  auto Parents = Context->getParents(*Usage.Expression);
580  if (Parents.size() == 1) {
581  if (const auto *Paren = Parents[0].get<ParenExpr>()) {
582  // Usage.Expression will be replaced with the new index variable,
583  // and parenthesis around a simple DeclRefExpr can always be
584  // removed.
585  Range = Paren->getSourceRange();
586  } else if (const auto *UOP = Parents[0].get<UnaryOperator>()) {
587  // If we are taking the address of the loop variable, then we must
588  // not use a copy, as it would mean taking the address of the loop's
589  // local index instead.
590  // FIXME: This won't catch cases where the address is taken outside
591  // of the loop's body (for instance, in a function that got the
592  // loop's index as a const reference parameter), or where we take
593  // the address of a member (like "&Arr[i].A.B.C").
594  if (UOP->getOpcode() == UO_AddrOf)
595  CanCopy = false;
596  }
597  }
598  } else {
599  // The Usage expression is only null in case of lambda captures (which
600  // are VarDecl). If the index is captured by value, add '&' to capture
601  // by reference instead.
602  ReplaceText =
603  Usage.Kind == Usage::UK_CaptureByCopy ? "&" + VarName : VarName;
604  }
605  TUInfo->getReplacedVars().insert(std::make_pair(Loop, IndexVar));
606  Diag << FixItHint::CreateReplacement(
607  CharSourceRange::getTokenRange(Range), ReplaceText);
608  }
609  }
610 
611  // Now, we need to construct the new range expression.
612  SourceRange ParenRange(Loop->getLParenLoc(), Loop->getRParenLoc());
613 
614  QualType Type = Context->getAutoDeductType();
615  if (!Descriptor.ElemType.isNull() && Descriptor.ElemType->isFundamentalType())
616  Type = Descriptor.ElemType.getUnqualifiedType();
617 
618  // If the new variable name is from the aliased variable, then the reference
619  // type for the new variable should only be used if the aliased variable was
620  // declared as a reference.
621  bool IsCheapToCopy =
622  !Descriptor.ElemType.isNull() &&
623  Descriptor.ElemType.isTriviallyCopyableType(*Context) &&
624  // TypeInfo::Width is in bits.
625  Context->getTypeInfo(Descriptor.ElemType).Width <= 8 * MaxCopySize;
626  bool UseCopy = CanCopy && ((VarNameFromAlias && !AliasVarIsRef) ||
627  (Descriptor.DerefByConstRef && IsCheapToCopy));
628 
629  if (!UseCopy) {
630  if (Descriptor.DerefByConstRef) {
631  Type = Context->getLValueReferenceType(Context->getConstType(Type));
632  } else if (Descriptor.DerefByValue) {
633  if (!IsCheapToCopy)
634  Type = Context->getRValueReferenceType(Type);
635  } else {
636  Type = Context->getLValueReferenceType(Type);
637  }
638  }
639 
640  StringRef MaybeDereference = Descriptor.ContainerNeedsDereference ? "*" : "";
641  std::string TypeString = Type.getAsString(getLangOpts());
642  std::string Range = ("(" + TypeString + " " + VarName + " : " +
643  MaybeDereference + Descriptor.ContainerString + ")")
644  .str();
645  Diag << FixItHint::CreateReplacement(
646  CharSourceRange::getTokenRange(ParenRange), Range);
647  TUInfo->getGeneratedDecls().insert(make_pair(Loop, VarName));
648 }
649 
650 /// \brief Returns a string which refers to the container iterated over.
651 StringRef LoopConvertCheck::getContainerString(ASTContext *Context,
652  const ForStmt *Loop,
653  const Expr *ContainerExpr) {
654  StringRef ContainerString;
655  if (isa<CXXThisExpr>(ContainerExpr->IgnoreParenImpCasts())) {
656  ContainerString = "this";
657  } else {
658  ContainerString =
659  getStringFromRange(Context->getSourceManager(), Context->getLangOpts(),
660  ContainerExpr->getSourceRange());
661  }
662 
663  return ContainerString;
664 }
665 
666 /// \brief Determines what kind of 'auto' must be used after converting a for
667 /// loop that iterates over an array or pseudoarray.
668 void LoopConvertCheck::getArrayLoopQualifiers(ASTContext *Context,
669  const BoundNodes &Nodes,
670  const Expr *ContainerExpr,
671  const UsageResult &Usages,
672  RangeDescriptor &Descriptor) {
673  // On arrays and pseudoarrays, we must figure out the qualifiers from the
674  // usages.
675  if (usagesAreConst(Context, Usages) ||
676  containerIsConst(ContainerExpr, Descriptor.ContainerNeedsDereference)) {
677  Descriptor.DerefByConstRef = true;
678  }
679  if (usagesReturnRValues(Usages)) {
680  // If the index usages (dereference, subscript, at, ...) return rvalues,
681  // then we should not use a reference, because we need to keep the code
682  // correct if it mutates the returned objects.
683  Descriptor.DerefByValue = true;
684  }
685  // Try to find the type of the elements on the container, to check if
686  // they are trivially copyable.
687  for (const Usage &U : Usages) {
688  if (!U.Expression || U.Expression->getType().isNull())
689  continue;
690  QualType Type = U.Expression->getType().getCanonicalType();
691  if (U.Kind == Usage::UK_MemberThroughArrow) {
692  if (!Type->isPointerType()) {
693  continue;
694  }
695  Type = Type->getPointeeType();
696  }
697  Descriptor.ElemType = Type;
698  }
699 }
700 
701 /// \brief Determines what kind of 'auto' must be used after converting an
702 /// iterator based for loop.
703 void LoopConvertCheck::getIteratorLoopQualifiers(ASTContext *Context,
704  const BoundNodes &Nodes,
705  RangeDescriptor &Descriptor) {
706  // The matchers for iterator loops provide bound nodes to obtain this
707  // information.
708  const auto *InitVar = Nodes.getNodeAs<VarDecl>(InitVarName);
709  QualType CanonicalInitVarType = InitVar->getType().getCanonicalType();
710  const auto *DerefByValueType =
711  Nodes.getNodeAs<QualType>(DerefByValueResultName);
712  Descriptor.DerefByValue = DerefByValueType;
713 
714  if (Descriptor.DerefByValue) {
715  // If the dereference operator returns by value then test for the
716  // canonical const qualification of the init variable type.
717  Descriptor.DerefByConstRef = CanonicalInitVarType.isConstQualified();
718  Descriptor.ElemType = *DerefByValueType;
719  } else {
720  if (const auto *DerefType =
721  Nodes.getNodeAs<QualType>(DerefByRefResultName)) {
722  // A node will only be bound with DerefByRefResultName if we're dealing
723  // with a user-defined iterator type. Test the const qualification of
724  // the reference type.
725  auto ValueType = DerefType->getNonReferenceType();
726 
727  Descriptor.DerefByConstRef = ValueType.isConstQualified();
728  Descriptor.ElemType = ValueType;
729  } else {
730  // By nature of the matcher this case is triggered only for built-in
731  // iterator types (i.e. pointers).
732  assert(isa<PointerType>(CanonicalInitVarType) &&
733  "Non-class iterator type is not a pointer type");
734 
735  // We test for const qualification of the pointed-at type.
736  Descriptor.DerefByConstRef =
737  CanonicalInitVarType->getPointeeType().isConstQualified();
738  Descriptor.ElemType = CanonicalInitVarType->getPointeeType();
739  }
740  }
741 }
742 
743 /// \brief Determines the parameters needed to build the range replacement.
744 void LoopConvertCheck::determineRangeDescriptor(
745  ASTContext *Context, const BoundNodes &Nodes, const ForStmt *Loop,
746  LoopFixerKind FixerKind, const Expr *ContainerExpr,
747  const UsageResult &Usages, RangeDescriptor &Descriptor) {
748  Descriptor.ContainerString = getContainerString(Context, Loop, ContainerExpr);
749 
750  if (FixerKind == LFK_Iterator)
751  getIteratorLoopQualifiers(Context, Nodes, Descriptor);
752  else
753  getArrayLoopQualifiers(Context, Nodes, ContainerExpr, Usages, Descriptor);
754 }
755 
756 /// \brief Check some of the conditions that must be met for the loop to be
757 /// convertible.
758 bool LoopConvertCheck::isConvertible(ASTContext *Context,
759  const ast_matchers::BoundNodes &Nodes,
760  const ForStmt *Loop,
761  LoopFixerKind FixerKind) {
762  // If we already modified the range of this for loop, don't do any further
763  // updates on this iteration.
764  if (TUInfo->getReplacedVars().count(Loop))
765  return false;
766 
767  // Check that we have exactly one index variable and at most one end variable.
768  const auto *LoopVar = Nodes.getNodeAs<VarDecl>(IncrementVarName);
769  const auto *CondVar = Nodes.getNodeAs<VarDecl>(ConditionVarName);
770  const auto *InitVar = Nodes.getNodeAs<VarDecl>(InitVarName);
771  if (!areSameVariable(LoopVar, CondVar) || !areSameVariable(LoopVar, InitVar))
772  return false;
773  const auto *EndVar = Nodes.getNodeAs<VarDecl>(EndVarName);
774  const auto *ConditionEndVar = Nodes.getNodeAs<VarDecl>(ConditionEndVarName);
775  if (EndVar && !areSameVariable(EndVar, ConditionEndVar))
776  return false;
777 
778  // FIXME: Try to put most of this logic inside a matcher.
779  if (FixerKind == LFK_Iterator) {
780  QualType InitVarType = InitVar->getType();
781  QualType CanonicalInitVarType = InitVarType.getCanonicalType();
782 
783  const auto *BeginCall = Nodes.getNodeAs<CXXMemberCallExpr>(BeginCallName);
784  assert(BeginCall && "Bad Callback. No begin call expression");
785  QualType CanonicalBeginType =
786  BeginCall->getMethodDecl()->getReturnType().getCanonicalType();
787  if (CanonicalBeginType->isPointerType() &&
788  CanonicalInitVarType->isPointerType()) {
789  // If the initializer and the variable are both pointers check if the
790  // un-qualified pointee types match, otherwise we don't use auto.
791  if (!Context->hasSameUnqualifiedType(
792  CanonicalBeginType->getPointeeType(),
793  CanonicalInitVarType->getPointeeType()))
794  return false;
795  } else if (!Context->hasSameType(CanonicalInitVarType,
796  CanonicalBeginType)) {
797  // Check for qualified types to avoid conversions from non-const to const
798  // iterator types.
799  return false;
800  }
801  } else if (FixerKind == LFK_PseudoArray) {
802  // This call is required to obtain the container.
803  const auto *EndCall = Nodes.getNodeAs<CXXMemberCallExpr>(EndCallName);
804  if (!EndCall || !dyn_cast<MemberExpr>(EndCall->getCallee()))
805  return false;
806  }
807  return true;
808 }
809 
810 void LoopConvertCheck::check(const MatchFinder::MatchResult &Result) {
811  const BoundNodes &Nodes = Result.Nodes;
812  Confidence ConfidenceLevel(Confidence::CL_Safe);
813  ASTContext *Context = Result.Context;
814 
815  const ForStmt *Loop;
816  LoopFixerKind FixerKind;
817  RangeDescriptor Descriptor;
818 
819  if ((Loop = Nodes.getNodeAs<ForStmt>(LoopNameArray))) {
820  FixerKind = LFK_Array;
821  } else if ((Loop = Nodes.getNodeAs<ForStmt>(LoopNameIterator))) {
822  FixerKind = LFK_Iterator;
823  } else {
824  Loop = Nodes.getNodeAs<ForStmt>(LoopNamePseudoArray);
825  assert(Loop && "Bad Callback. No for statement");
826  FixerKind = LFK_PseudoArray;
827  }
828 
829  if (!isConvertible(Context, Nodes, Loop, FixerKind))
830  return;
831 
832  const auto *LoopVar = Nodes.getNodeAs<VarDecl>(IncrementVarName);
833  const auto *EndVar = Nodes.getNodeAs<VarDecl>(EndVarName);
834 
835  // If the loop calls end()/size() after each iteration, lower our confidence
836  // level.
837  if (FixerKind != LFK_Array && !EndVar)
838  ConfidenceLevel.lowerTo(Confidence::CL_Reasonable);
839 
840  // If the end comparison isn't a variable, we can try to work with the
841  // expression the loop variable is being tested against instead.
842  const auto *EndCall = Nodes.getNodeAs<CXXMemberCallExpr>(EndCallName);
843  const auto *BoundExpr = Nodes.getNodeAs<Expr>(ConditionBoundName);
844 
845  // Find container expression of iterators and pseudoarrays, and determine if
846  // this expression needs to be dereferenced to obtain the container.
847  // With array loops, the container is often discovered during the
848  // ForLoopIndexUseVisitor traversal.
849  const Expr *ContainerExpr = nullptr;
850  if (FixerKind == LFK_Iterator) {
851  ContainerExpr = findContainer(Context, LoopVar->getInit(),
852  EndVar ? EndVar->getInit() : EndCall,
853  &Descriptor.ContainerNeedsDereference);
854  } else if (FixerKind == LFK_PseudoArray) {
855  ContainerExpr = EndCall->getImplicitObjectArgument();
856  Descriptor.ContainerNeedsDereference =
857  dyn_cast<MemberExpr>(EndCall->getCallee())->isArrow();
858  }
859 
860  // We must know the container or an array length bound.
861  if (!ContainerExpr && !BoundExpr)
862  return;
863 
864  ForLoopIndexUseVisitor Finder(Context, LoopVar, EndVar, ContainerExpr,
865  BoundExpr,
866  Descriptor.ContainerNeedsDereference);
867 
868  // Find expressions and variables on which the container depends.
869  if (ContainerExpr) {
870  ComponentFinderASTVisitor ComponentFinder;
871  ComponentFinder.findExprComponents(ContainerExpr->IgnoreParenImpCasts());
872  Finder.addComponents(ComponentFinder.getComponents());
873  }
874 
875  // Find usages of the loop index. If they are not used in a convertible way,
876  // stop here.
877  if (!Finder.findAndVerifyUsages(Loop->getBody()))
878  return;
879  ConfidenceLevel.lowerTo(Finder.getConfidenceLevel());
880 
881  // Obtain the container expression, if we don't have it yet.
882  if (FixerKind == LFK_Array) {
883  ContainerExpr = Finder.getContainerIndexed()->IgnoreParenImpCasts();
884 
885  // Very few loops are over expressions that generate arrays rather than
886  // array variables. Consider loops over arrays that aren't just represented
887  // by a variable to be risky conversions.
888  if (!getReferencedVariable(ContainerExpr) &&
889  !isDirectMemberExpr(ContainerExpr))
890  ConfidenceLevel.lowerTo(Confidence::CL_Risky);
891  }
892 
893  // Find out which qualifiers we have to use in the loop range.
894  const UsageResult &Usages = Finder.getUsages();
895  determineRangeDescriptor(Context, Nodes, Loop, FixerKind, ContainerExpr,
896  Usages, Descriptor);
897 
898  // Ensure that we do not try to move an expression dependent on a local
899  // variable declared inside the loop outside of it.
900  // FIXME: Determine when the external dependency isn't an expression converted
901  // by another loop.
902  TUInfo->getParentFinder().gatherAncestors(Context->getTranslationUnitDecl());
903  DependencyFinderASTVisitor DependencyFinder(
904  &TUInfo->getParentFinder().getStmtToParentStmtMap(),
905  &TUInfo->getParentFinder().getDeclToParentStmtMap(),
906  &TUInfo->getReplacedVars(), Loop);
907 
908  if (DependencyFinder.dependsOnInsideVariable(ContainerExpr) ||
909  Descriptor.ContainerString.empty() || Usages.empty() ||
910  ConfidenceLevel.getLevel() < MinConfidence)
911  return;
912 
913  doConversion(Context, LoopVar, getReferencedVariable(ContainerExpr), Usages,
914  Finder.getAliasDecl(), Finder.aliasUseRequired(),
915  Finder.aliasFromForInit(), Loop, Descriptor);
916 }
917 
918 } // namespace modernize
919 } // namespace tidy
920 } // namespace clang
static const char DerefByRefResultName[]
Confidence::Level getConfidenceLevel() const
Accessor for ConfidenceLevel.
Discover usages of expressions consisting of index or iterator access.
void store(ClangTidyOptions::OptionMap &Options, StringRef LocalName, StringRef Value) const
Stores an option with the check-local name LocalName with string value Value to Options.
Definition: ClangTidy.cpp:449
StatementMatcher makeIteratorLoopMatcher()
The matcher used for iterator-based for loops.
void storeOptions(ClangTidyOptions::OptionMap &Opts) override
Should store all options supported by this check with their current values or default values for opti...
static const char ConditionVarName[]
static const Expr * getContainerFromBeginEndCall(const Expr *Init, bool IsBegin, bool *IsArrow)
Determine whether Init appears to be an initializing an iterator.
StatementMatcher makeArrayLoopMatcher()
The matcher for loops over arrays.
llvm::SmallVector< Usage, 8 > UsageResult
StringHandle Name
LangOptions getLangOpts() const
Returns the language options from the context.
Definition: ClangTidy.h:187
static const char EndCallName[]
A class to encapsulate lowering of the tool&#39;s confidence level.
const Expr * getContainerIndexed() const
Get the container indexed by IndexVar, if any.
static const StatementMatcher IntegerComparisonMatcher
static const DeclarationMatcher InitToZeroMatcher
Class used to determine if an expression is dependent on a variable declared inside of the loop where...
Base class for all clang-tidy checks.
Definition: ClangTidy.h:127
bool aliasUseRequired() const
Indicates if the alias declaration was in a place where it cannot simply be removed but rather replac...
const Expr * digThroughConstructors(const Expr *E)
Look through conversion/copy constructors to find the explicit initialization expression, returning it is found.
static const Expr * findContainer(ASTContext *Context, const Expr *BeginExpr, const Expr *EndExpr, bool *ContainerNeedsDereference)
Determines the container whose begin() and end() functions are called for an iterator-based loop...
static const char InitVarName[]
static const ValueDecl * getReferencedVariable(const Expr *E)
If the given expression is actually a DeclRefExpr or a MemberExpr, find and return the underlying Val...
static const char EndVarName[]
bool aliasFromForInit() const
Indicates if the alias declaration came from the init clause of a nested for loop.
void registerMatchers(ast_matchers::MatchFinder *Finder) override
Override this to register AST matchers with Finder.
const DeclRefExpr * getDeclRef(const Expr *E)
Returns the DeclRefExpr represented by E, or NULL if there isn&#39;t one.
void findExprComponents(const clang::Expr *SourceExpr)
Find the components of an expression and place them in a ComponentVector.
std::string createIndexName()
Generate a new index name.
const ComponentVector & getComponents()
Accessor for Components.
Level getLevel() const
Return the internal confidence level.
bool areSameVariable(const ValueDecl *First, const ValueDecl *Second)
Returns true when two ValueDecls are the same variable.
static bool usagesReturnRValues(const UsageResult &Usages)
Returns true if the elements of the container are never accessed by reference.
static const char ConditionBoundName[]
static const StatementMatcher IncrementVarMatcher
Create names for generated variables within a particular statement.
static bool usagesAreConst(ASTContext *Context, const UsageResult &Usages)
Returns true when it can be guaranteed that the elements of the container are not being modified...
static const char IncrementVarName[]
bool areSameExpr(ASTContext *Context, const Expr *First, const Expr *Second)
Returns true when two Exprs are equivalent.
static bool canBeModified(ASTContext *Context, const Expr *E)
Given an expression that represents an usage of an element from the containter that we are iterating ...
std::map< std::string, std::string > OptionMap
static const char ConditionEndVarName[]
static const char LoopNameArray[]
static StringRef getStringFromRange(SourceManager &SourceMgr, const LangOptions &LangOpts, SourceRange Range)
Obtain the original source code text from a SourceRange.
static const char DerefByValueResultName[]
void check(const ast_matchers::MatchFinder::MatchResult &Result) override
ClangTidyChecks that register ASTMatchers should do the actual work in here.
static bool isDirectMemberExpr(const Expr *E)
Returns true when the given expression is a member expression whose base is this (implicitly or not)...
static const char LoopNamePseudoArray[]
bool findAndVerifyUsages(const Stmt *Body)
Finds all uses of IndexVar in Body, placing all usages in Usages, and returns true if IndexVar was on...
The information needed to describe a valid convertible usage of an array index or iterator...
CharSourceRange Range
SourceRange for the file name.
void addComponents(const ComponentVector &Components)
Add a set of components that we should consider relevant to the container.
void lowerTo(Confidence::Level Level)
Lower the internal confidence level to Level, but do not raise it.
static const char BeginCallName[]
Every ClangTidyCheck reports errors through a DiagnosticsEngine provided by this context.
static const TypeMatcher AnyType
Class used to find the variables and member expressions on which an arbitrary expression depends...
static const char LoopNameIterator[]
static bool containerIsConst(const Expr *ContainerExpr, bool Dereference)
Returns true if the container is const-qualified.
StatementMatcher makePseudoArrayLoopMatcher()
The matcher used for array-like containers (pseudoarrays).
DiagnosticBuilder diag(SourceLocation Loc, StringRef Description, DiagnosticIDs::Level Level=DiagnosticIDs::Warning)
Add a diagnostic with the check&#39;s name.
Definition: ClangTidy.cpp:416
const UsageResult & getUsages() const
Accessor for Usages.
const DeclStmt * getAliasDecl() const
Returns the statement declaring the variable created as an alias for the loop element, if any.