clang-tools  3.9.0
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 "../utils/Matchers.h"
12 #include "clang/AST/ASTContext.h"
13 #include "clang/ASTMatchers/ASTMatchFinder.h"
14 
15 using namespace clang;
16 using namespace clang::ast_matchers;
17 using namespace llvm;
18 
19 namespace clang {
20 namespace tidy {
21 namespace modernize {
22 
23 static const char LoopNameArray[] = "forLoopArray";
24 static const char LoopNameIterator[] = "forLoopIterator";
25 static const char LoopNamePseudoArray[] = "forLoopPseudoArray";
26 static const char ConditionBoundName[] = "conditionBound";
27 static const char ConditionVarName[] = "conditionVar";
28 static const char IncrementVarName[] = "incrementVar";
29 static const char InitVarName[] = "initVar";
30 static const char BeginCallName[] = "beginCall";
31 static const char EndCallName[] = "endCall";
32 static const char ConditionEndVarName[] = "conditionEndVar";
33 static const char EndVarName[] = "endVar";
34 static const char DerefByValueResultName[] = "derefByValueResult";
35 static const char DerefByRefResultName[] = "derefByRefResult";
36 
37 // shared matchers
38 static const TypeMatcher AnyType = anything();
39 
40 static const StatementMatcher IntegerComparisonMatcher =
41  expr(ignoringParenImpCasts(
42  declRefExpr(to(varDecl(hasType(isInteger())).bind(ConditionVarName)))));
43 
44 static const DeclarationMatcher InitToZeroMatcher =
45  varDecl(hasInitializer(ignoringParenImpCasts(integerLiteral(equals(0)))))
46  .bind(InitVarName);
47 
48 static const StatementMatcher IncrementVarMatcher =
49  declRefExpr(to(varDecl(hasType(isInteger())).bind(IncrementVarName)));
50 
51 /// \brief The matcher for loops over arrays.
52 ///
53 /// In this general example, assuming 'j' and 'k' are of integral type:
54 /// \code
55 /// for (int i = 0; j < 3 + 2; ++k) { ... }
56 /// \endcode
57 /// The following string identifiers are bound to these parts of the AST:
58 /// ConditionVarName: 'j' (as a VarDecl)
59 /// ConditionBoundName: '3 + 2' (as an Expr)
60 /// InitVarName: 'i' (as a VarDecl)
61 /// IncrementVarName: 'k' (as a VarDecl)
62 /// LoopName: The entire for loop (as a ForStmt)
63 ///
64 /// Client code will need to make sure that:
65 /// - The three index variables identified by the matcher are the same
66 /// VarDecl.
67 /// - The index variable is only used as an array index.
68 /// - All arrays indexed by the loop are the same.
69 StatementMatcher makeArrayLoopMatcher() {
70  StatementMatcher ArrayBoundMatcher =
71  expr(hasType(isInteger())).bind(ConditionBoundName);
72 
73  return forStmt(
74  unless(isInTemplateInstantiation()),
75  hasLoopInit(declStmt(hasSingleDecl(InitToZeroMatcher))),
76  hasCondition(anyOf(
77  binaryOperator(hasOperatorName("<"),
79  hasRHS(ArrayBoundMatcher)),
80  binaryOperator(hasOperatorName(">"), hasLHS(ArrayBoundMatcher),
81  hasRHS(IntegerComparisonMatcher)))),
82  hasIncrement(unaryOperator(hasOperatorName("++"),
83  hasUnaryOperand(IncrementVarMatcher))))
84  .bind(LoopNameArray);
85 }
86 
87 /// \brief The matcher used for iterator-based for loops.
88 ///
89 /// This matcher is more flexible than array-based loops. It will match
90 /// catch loops of the following textual forms (regardless of whether the
91 /// iterator type is actually a pointer type or a class type):
92 ///
93 /// Assuming f, g, and h are of type containerType::iterator,
94 /// \code
95 /// for (containerType::iterator it = container.begin(),
96 /// e = createIterator(); f != g; ++h) { ... }
97 /// for (containerType::iterator it = container.begin();
98 /// f != anotherContainer.end(); ++h) { ... }
99 /// \endcode
100 /// The following string identifiers are bound to the parts of the AST:
101 /// InitVarName: 'it' (as a VarDecl)
102 /// ConditionVarName: 'f' (as a VarDecl)
103 /// LoopName: The entire for loop (as a ForStmt)
104 /// In the first example only:
105 /// EndVarName: 'e' (as a VarDecl)
106 /// ConditionEndVarName: 'g' (as a VarDecl)
107 /// In the second example only:
108 /// EndCallName: 'container.end()' (as a CXXMemberCallExpr)
109 ///
110 /// Client code will need to make sure that:
111 /// - The iterator variables 'it', 'f', and 'h' are the same.
112 /// - The two containers on which 'begin' and 'end' are called are the same.
113 /// - If the end iterator variable 'g' is defined, it is the same as 'f'.
114 StatementMatcher makeIteratorLoopMatcher() {
115  StatementMatcher BeginCallMatcher =
116  cxxMemberCallExpr(
117  argumentCountIs(0),
118  callee(cxxMethodDecl(anyOf(hasName("begin"), hasName("cbegin")))))
119  .bind(BeginCallName);
120 
121  DeclarationMatcher InitDeclMatcher =
122  varDecl(hasInitializer(anyOf(ignoringParenImpCasts(BeginCallMatcher),
123  materializeTemporaryExpr(
124  ignoringParenImpCasts(BeginCallMatcher)),
125  hasDescendant(BeginCallMatcher))))
126  .bind(InitVarName);
127 
128  DeclarationMatcher EndDeclMatcher =
129  varDecl(hasInitializer(anything())).bind(EndVarName);
130 
131  StatementMatcher EndCallMatcher = cxxMemberCallExpr(
132  argumentCountIs(0),
133  callee(cxxMethodDecl(anyOf(hasName("end"), hasName("cend")))));
134 
135  StatementMatcher IteratorBoundMatcher =
136  expr(anyOf(ignoringParenImpCasts(
137  declRefExpr(to(varDecl().bind(ConditionEndVarName)))),
138  ignoringParenImpCasts(expr(EndCallMatcher).bind(EndCallName)),
139  materializeTemporaryExpr(ignoringParenImpCasts(
140  expr(EndCallMatcher).bind(EndCallName)))));
141 
142  StatementMatcher IteratorComparisonMatcher = expr(
143  ignoringParenImpCasts(declRefExpr(to(varDecl().bind(ConditionVarName)))));
144 
145  auto OverloadedNEQMatcher = ignoringImplicit(
146  cxxOperatorCallExpr(hasOverloadedOperatorName("!="), argumentCountIs(2),
147  hasArgument(0, IteratorComparisonMatcher),
148  hasArgument(1, IteratorBoundMatcher)));
149 
150  // This matcher tests that a declaration is a CXXRecordDecl that has an
151  // overloaded operator*(). If the operator*() returns by value instead of by
152  // reference then the return type is tagged with DerefByValueResultName.
153  internal::Matcher<VarDecl> TestDerefReturnsByValue =
154  hasType(cxxRecordDecl(hasMethod(allOf(
155  hasOverloadedOperatorName("*"),
156  anyOf(
157  // Tag the return type if it's by value.
158  returns(qualType(unless(hasCanonicalType(referenceType())))
159  .bind(DerefByValueResultName)),
160  returns(
161  // Skip loops where the iterator's operator* returns an
162  // rvalue reference. This is just weird.
163  qualType(unless(hasCanonicalType(rValueReferenceType())))
164  .bind(DerefByRefResultName)))))));
165 
166  return forStmt(
167  unless(isInTemplateInstantiation()),
168  hasLoopInit(anyOf(declStmt(declCountIs(2),
169  containsDeclaration(0, InitDeclMatcher),
170  containsDeclaration(1, EndDeclMatcher)),
171  declStmt(hasSingleDecl(InitDeclMatcher)))),
172  hasCondition(
173  anyOf(binaryOperator(hasOperatorName("!="),
174  hasLHS(IteratorComparisonMatcher),
175  hasRHS(IteratorBoundMatcher)),
176  binaryOperator(hasOperatorName("!="),
177  hasLHS(IteratorBoundMatcher),
178  hasRHS(IteratorComparisonMatcher)),
179  OverloadedNEQMatcher)),
180  hasIncrement(anyOf(
181  unaryOperator(hasOperatorName("++"),
182  hasUnaryOperand(declRefExpr(
183  to(varDecl(hasType(pointsTo(AnyType)))
184  .bind(IncrementVarName))))),
185  cxxOperatorCallExpr(
186  hasOverloadedOperatorName("++"),
187  hasArgument(
188  0, declRefExpr(to(varDecl(TestDerefReturnsByValue)
189  .bind(IncrementVarName))))))))
190  .bind(LoopNameIterator);
191 }
192 
193 /// \brief The matcher used for array-like containers (pseudoarrays).
194 ///
195 /// This matcher is more flexible than array-based loops. It will match
196 /// loops of the following textual forms (regardless of whether the
197 /// iterator type is actually a pointer type or a class type):
198 ///
199 /// Assuming f, g, and h are of type containerType::iterator,
200 /// \code
201 /// for (int i = 0, j = container.size(); f < g; ++h) { ... }
202 /// for (int i = 0; f < container.size(); ++h) { ... }
203 /// \endcode
204 /// The following string identifiers are bound to the parts of the AST:
205 /// InitVarName: 'i' (as a VarDecl)
206 /// ConditionVarName: 'f' (as a VarDecl)
207 /// LoopName: The entire for loop (as a ForStmt)
208 /// In the first example only:
209 /// EndVarName: 'j' (as a VarDecl)
210 /// ConditionEndVarName: 'g' (as a VarDecl)
211 /// In the second example only:
212 /// EndCallName: 'container.size()' (as a CXXMemberCallExpr)
213 ///
214 /// Client code will need to make sure that:
215 /// - The index variables 'i', 'f', and 'h' are the same.
216 /// - The containers on which 'size()' is called is the container indexed.
217 /// - The index variable is only used in overloaded operator[] or
218 /// container.at().
219 /// - If the end iterator variable 'g' is defined, it is the same as 'j'.
220 /// - The container's iterators would not be invalidated during the loop.
221 StatementMatcher makePseudoArrayLoopMatcher() {
222  // Test that the incoming type has a record declaration that has methods
223  // called 'begin' and 'end'. If the incoming type is const, then make sure
224  // these methods are also marked const.
225  //
226  // FIXME: To be completely thorough this matcher should also ensure the
227  // return type of begin/end is an iterator that dereferences to the same as
228  // what operator[] or at() returns. Such a test isn't likely to fail except
229  // for pathological cases.
230  //
231  // FIXME: Also, a record doesn't necessarily need begin() and end(). Free
232  // functions called begin() and end() taking the container as an argument
233  // are also allowed.
234  TypeMatcher RecordWithBeginEnd = qualType(anyOf(
235  qualType(isConstQualified(),
236  hasDeclaration(cxxRecordDecl(
237  hasMethod(cxxMethodDecl(hasName("begin"), isConst())),
238  hasMethod(cxxMethodDecl(hasName("end"),
239  isConst())))) // hasDeclaration
240  ), // qualType
241  qualType(
242  unless(isConstQualified()),
243  hasDeclaration(cxxRecordDecl(hasMethod(hasName("begin")),
244  hasMethod(hasName("end"))))) // qualType
245  ));
246 
247  StatementMatcher SizeCallMatcher = cxxMemberCallExpr(
248  argumentCountIs(0),
249  callee(cxxMethodDecl(anyOf(hasName("size"), hasName("length")))),
250  on(anyOf(hasType(pointsTo(RecordWithBeginEnd)),
251  hasType(RecordWithBeginEnd))));
252 
253  StatementMatcher EndInitMatcher =
254  expr(anyOf(ignoringParenImpCasts(expr(SizeCallMatcher).bind(EndCallName)),
255  explicitCastExpr(hasSourceExpression(ignoringParenImpCasts(
256  expr(SizeCallMatcher).bind(EndCallName))))));
257 
258  DeclarationMatcher EndDeclMatcher =
259  varDecl(hasInitializer(EndInitMatcher)).bind(EndVarName);
260 
261  StatementMatcher IndexBoundMatcher =
262  expr(anyOf(ignoringParenImpCasts(declRefExpr(to(
263  varDecl(hasType(isInteger())).bind(ConditionEndVarName)))),
264  EndInitMatcher));
265 
266  return forStmt(
267  unless(isInTemplateInstantiation()),
268  hasLoopInit(
269  anyOf(declStmt(declCountIs(2),
270  containsDeclaration(0, InitToZeroMatcher),
271  containsDeclaration(1, EndDeclMatcher)),
272  declStmt(hasSingleDecl(InitToZeroMatcher)))),
273  hasCondition(anyOf(
274  binaryOperator(hasOperatorName("<"),
275  hasLHS(IntegerComparisonMatcher),
276  hasRHS(IndexBoundMatcher)),
277  binaryOperator(hasOperatorName(">"), hasLHS(IndexBoundMatcher),
278  hasRHS(IntegerComparisonMatcher)))),
279  hasIncrement(unaryOperator(hasOperatorName("++"),
280  hasUnaryOperand(IncrementVarMatcher))))
281  .bind(LoopNamePseudoArray);
282 }
283 
284 /// \brief Determine whether Init appears to be an initializing an iterator.
285 ///
286 /// If it is, returns the object whose begin() or end() method is called, and
287 /// the output parameter isArrow is set to indicate whether the initialization
288 /// is called via . or ->.
289 static const Expr *getContainerFromBeginEndCall(const Expr *Init, bool IsBegin,
290  bool *IsArrow) {
291  // FIXME: Maybe allow declaration/initialization outside of the for loop.
292  const auto *TheCall =
293  dyn_cast_or_null<CXXMemberCallExpr>(digThroughConstructors(Init));
294  if (!TheCall || TheCall->getNumArgs() != 0)
295  return nullptr;
296 
297  const auto *Member = dyn_cast<MemberExpr>(TheCall->getCallee());
298  if (!Member)
299  return nullptr;
300  StringRef Name = Member->getMemberDecl()->getName();
301  StringRef TargetName = IsBegin ? "begin" : "end";
302  StringRef ConstTargetName = IsBegin ? "cbegin" : "cend";
303  if (Name != TargetName && Name != ConstTargetName)
304  return nullptr;
305 
306  const Expr *SourceExpr = Member->getBase();
307  if (!SourceExpr)
308  return nullptr;
309 
310  *IsArrow = Member->isArrow();
311  return SourceExpr;
312 }
313 
314 /// \brief Determines the container whose begin() and end() functions are called
315 /// for an iterator-based loop.
316 ///
317 /// BeginExpr must be a member call to a function named "begin()", and EndExpr
318 /// must be a member.
319 static const Expr *findContainer(ASTContext *Context, const Expr *BeginExpr,
320  const Expr *EndExpr,
321  bool *ContainerNeedsDereference) {
322  // Now that we know the loop variable and test expression, make sure they are
323  // valid.
324  bool BeginIsArrow = false;
325  bool EndIsArrow = false;
326  const Expr *BeginContainerExpr =
327  getContainerFromBeginEndCall(BeginExpr, /*IsBegin=*/true, &BeginIsArrow);
328  if (!BeginContainerExpr)
329  return nullptr;
330 
331  const Expr *EndContainerExpr =
332  getContainerFromBeginEndCall(EndExpr, /*IsBegin=*/false, &EndIsArrow);
333  // Disallow loops that try evil things like this (note the dot and arrow):
334  // for (IteratorType It = Obj.begin(), E = Obj->end(); It != E; ++It) { }
335  if (!EndContainerExpr || BeginIsArrow != EndIsArrow ||
336  !areSameExpr(Context, EndContainerExpr, BeginContainerExpr))
337  return nullptr;
338 
339  *ContainerNeedsDereference = BeginIsArrow;
340  return BeginContainerExpr;
341 }
342 
343 /// \brief Obtain the original source code text from a SourceRange.
344 static StringRef getStringFromRange(SourceManager &SourceMgr,
345  const LangOptions &LangOpts,
346  SourceRange Range) {
347  if (SourceMgr.getFileID(Range.getBegin()) !=
348  SourceMgr.getFileID(Range.getEnd())) {
349  return StringRef(); // Empty string.
350  }
351 
352  return Lexer::getSourceText(CharSourceRange(Range, true), SourceMgr,
353  LangOpts);
354 }
355 
356 /// \brief If the given expression is actually a DeclRefExpr or a MemberExpr,
357 /// find and return the underlying ValueDecl; otherwise, return NULL.
358 static const ValueDecl *getReferencedVariable(const Expr *E) {
359  if (const DeclRefExpr *DRE = getDeclRef(E))
360  return dyn_cast<VarDecl>(DRE->getDecl());
361  if (const auto *Mem = dyn_cast<MemberExpr>(E->IgnoreParenImpCasts()))
362  return dyn_cast<FieldDecl>(Mem->getMemberDecl());
363  return nullptr;
364 }
365 
366 /// \brief Returns true when the given expression is a member expression
367 /// whose base is `this` (implicitly or not).
368 static bool isDirectMemberExpr(const Expr *E) {
369  if (const auto *Member = dyn_cast<MemberExpr>(E->IgnoreParenImpCasts()))
370  return isa<CXXThisExpr>(Member->getBase()->IgnoreParenImpCasts());
371  return false;
372 }
373 
374 /// \brief Given an expression that represents an usage of an element from the
375 /// containter that we are iterating over, returns false when it can be
376 /// guaranteed this element cannot be modified as a result of this usage.
377 static bool canBeModified(ASTContext *Context, const Expr *E) {
378  if (E->getType().isConstQualified())
379  return false;
380  auto Parents = Context->getParents(*E);
381  if (Parents.size() != 1)
382  return true;
383  if (const auto *Cast = Parents[0].get<ImplicitCastExpr>()) {
384  if ((Cast->getCastKind() == CK_NoOp &&
385  Cast->getType() == E->getType().withConst()) ||
386  (Cast->getCastKind() == CK_LValueToRValue &&
387  !Cast->getType().isNull() && Cast->getType()->isFundamentalType()))
388  return false;
389  }
390  // FIXME: Make this function more generic.
391  return true;
392 }
393 
394 /// \brief Returns true when it can be guaranteed that the elements of the
395 /// container are not being modified.
396 static bool usagesAreConst(ASTContext *Context, const UsageResult &Usages) {
397  for (const Usage &U : Usages) {
398  // Lambda captures are just redeclarations (VarDecl) of the same variable,
399  // not expressions. If we want to know if a variable that is captured by
400  // reference can be modified in an usage inside the lambda's body, we need
401  // to find the expression corresponding to that particular usage, later in
402  // this loop.
403  if (U.Kind != Usage::UK_CaptureByCopy && U.Kind != Usage::UK_CaptureByRef &&
404  canBeModified(Context, U.Expression))
405  return false;
406  }
407  return true;
408 }
409 
410 /// \brief Returns true if the elements of the container are never accessed
411 /// by reference.
412 static bool usagesReturnRValues(const UsageResult &Usages) {
413  for (const auto &U : Usages) {
414  if (U.Expression && !U.Expression->isRValue())
415  return false;
416  }
417  return true;
418 }
419 
420 /// \brief Returns true if the container is const-qualified.
421 static bool containerIsConst(const Expr *ContainerExpr, bool Dereference) {
422  if (const auto *VDec = getReferencedVariable(ContainerExpr)) {
423  QualType CType = VDec->getType();
424  if (Dereference) {
425  if (!CType->isPointerType())
426  return false;
427  CType = CType->getPointeeType();
428  }
429  // If VDec is a reference to a container, Dereference is false,
430  // but we still need to check the const-ness of the underlying container
431  // type.
432  CType = CType.getNonReferenceType();
433  return CType.isConstQualified();
434  }
435  return false;
436 }
437 
438 LoopConvertCheck::RangeDescriptor::RangeDescriptor()
439  : ContainerNeedsDereference(false), DerefByConstRef(false),
440  DerefByValue(false) {}
441 
442 LoopConvertCheck::LoopConvertCheck(StringRef Name, ClangTidyContext *Context)
443  : ClangTidyCheck(Name, Context), TUInfo(new TUTrackingInfo),
444  MaxCopySize(std::stoull(Options.get("MaxCopySize", "16"))),
445  MinConfidence(StringSwitch<Confidence::Level>(
446  Options.get("MinConfidence", "reasonable"))
447  .Case("safe", Confidence::CL_Safe)
448  .Case("risky", Confidence::CL_Risky)
449  .Default(Confidence::CL_Reasonable)),
450  NamingStyle(StringSwitch<VariableNamer::NamingStyle>(
451  Options.get("NamingStyle", "CamelCase"))
452  .Case("camelBack", VariableNamer::NS_CamelBack)
453  .Case("lower_case", VariableNamer::NS_LowerCase)
454  .Case("UPPER_CASE", VariableNamer::NS_UpperCase)
455  .Default(VariableNamer::NS_CamelCase)) {}
456 
458  Options.store(Opts, "MaxCopySize", std::to_string(MaxCopySize));
459  SmallVector<std::string, 3> Confs{"risky", "reasonable", "safe"};
460  Options.store(Opts, "MinConfidence", Confs[static_cast<int>(MinConfidence)]);
461 
462  SmallVector<std::string, 4> Styles{"camelBack", "CamelCase", "lower_case",
463  "UPPER_CASE"};
464  Options.store(Opts, "NamingStyle", Styles[static_cast<int>(NamingStyle)]);
465 }
466 
468  // Only register the matchers for C++. Because this checker is used for
469  // modernization, it is reasonable to run it on any C++ standard with the
470  // assumption the user is trying to modernize their codebase.
471  if (!getLangOpts().CPlusPlus)
472  return;
473 
474  Finder->addMatcher(makeArrayLoopMatcher(), this);
475  Finder->addMatcher(makeIteratorLoopMatcher(), this);
476  Finder->addMatcher(makePseudoArrayLoopMatcher(), this);
477 }
478 
479 /// \brief Given the range of a single declaration, such as:
480 /// \code
481 /// unsigned &ThisIsADeclarationThatCanSpanSeveralLinesOfCode =
482 /// InitializationValues[I];
483 /// next_instruction;
484 /// \endcode
485 /// Finds the range that has to be erased to remove this declaration without
486 /// leaving empty lines, by extending the range until the beginning of the
487 /// next instruction.
488 ///
489 /// We need to delete a potential newline after the deleted alias, as
490 /// clang-format will leave empty lines untouched. For all other formatting we
491 /// rely on clang-format to fix it.
492 void LoopConvertCheck::getAliasRange(SourceManager &SM, SourceRange &Range) {
493  bool Invalid = false;
494  const char *TextAfter =
495  SM.getCharacterData(Range.getEnd().getLocWithOffset(1), &Invalid);
496  if (Invalid)
497  return;
498  unsigned Offset = std::strspn(TextAfter, " \t\r\n");
499  Range =
500  SourceRange(Range.getBegin(), Range.getEnd().getLocWithOffset(Offset));
501 }
502 
503 /// \brief Computes the changes needed to convert a given for loop, and
504 /// applies them.
505 void LoopConvertCheck::doConversion(
506  ASTContext *Context, const VarDecl *IndexVar,
507  const ValueDecl *MaybeContainer, const UsageResult &Usages,
508  const DeclStmt *AliasDecl, bool AliasUseRequired, bool AliasFromForInit,
509  const ForStmt *Loop, RangeDescriptor Descriptor) {
510  auto Diag = diag(Loop->getForLoc(), "use range-based for loop instead");
511 
512  std::string VarName;
513  bool VarNameFromAlias = (Usages.size() == 1) && AliasDecl;
514  bool AliasVarIsRef = false;
515  bool CanCopy = true;
516 
517  if (VarNameFromAlias) {
518  const auto *AliasVar = cast<VarDecl>(AliasDecl->getSingleDecl());
519  VarName = AliasVar->getName().str();
520  AliasVarIsRef = AliasVar->getType()->isReferenceType();
521 
522  // We keep along the entire DeclStmt to keep the correct range here.
523  SourceRange ReplaceRange = AliasDecl->getSourceRange();
524 
525  std::string ReplacementText;
526  if (AliasUseRequired) {
527  ReplacementText = VarName;
528  } else if (AliasFromForInit) {
529  // FIXME: Clang includes the location of the ';' but only for DeclStmt's
530  // in a for loop's init clause. Need to put this ';' back while removing
531  // the declaration of the alias variable. This is probably a bug.
532  ReplacementText = ";";
533  } else {
534  // Avoid leaving empty lines or trailing whitespaces.
535  getAliasRange(Context->getSourceManager(), ReplaceRange);
536  }
537 
538  Diag << FixItHint::CreateReplacement(
539  CharSourceRange::getTokenRange(ReplaceRange), ReplacementText);
540  // No further replacements are made to the loop, since the iterator or index
541  // was used exactly once - in the initialization of AliasVar.
542  } else {
543  VariableNamer Namer(&TUInfo->getGeneratedDecls(),
544  &TUInfo->getParentFinder().getStmtToParentStmtMap(),
545  Loop, IndexVar, MaybeContainer, Context, NamingStyle);
546  VarName = Namer.createIndexName();
547  // First, replace all usages of the array subscript expression with our new
548  // variable.
549  for (const auto &Usage : Usages) {
550  std::string ReplaceText;
551  SourceRange Range = Usage.Range;
552  if (Usage.Expression) {
553  // If this is an access to a member through the arrow operator, after
554  // the replacement it must be accessed through the '.' operator.
555  ReplaceText = Usage.Kind == Usage::UK_MemberThroughArrow ? VarName + "."
556  : VarName;
557  auto Parents = Context->getParents(*Usage.Expression);
558  if (Parents.size() == 1) {
559  if (const auto *Paren = Parents[0].get<ParenExpr>()) {
560  // Usage.Expression will be replaced with the new index variable,
561  // and parenthesis around a simple DeclRefExpr can always be
562  // removed.
563  Range = Paren->getSourceRange();
564  } else if (const auto *UOP = Parents[0].get<UnaryOperator>()) {
565  // If we are taking the address of the loop variable, then we must
566  // not use a copy, as it would mean taking the address of the loop's
567  // local index instead.
568  // FIXME: This won't catch cases where the address is taken outside
569  // of the loop's body (for instance, in a function that got the
570  // loop's index as a const reference parameter), or where we take
571  // the address of a member (like "&Arr[i].A.B.C").
572  if (UOP->getOpcode() == UO_AddrOf)
573  CanCopy = false;
574  }
575  }
576  } else {
577  // The Usage expression is only null in case of lambda captures (which
578  // are VarDecl). If the index is captured by value, add '&' to capture
579  // by reference instead.
580  ReplaceText =
581  Usage.Kind == Usage::UK_CaptureByCopy ? "&" + VarName : VarName;
582  }
583  TUInfo->getReplacedVars().insert(std::make_pair(Loop, IndexVar));
584  Diag << FixItHint::CreateReplacement(
585  CharSourceRange::getTokenRange(Range), ReplaceText);
586  }
587  }
588 
589  // Now, we need to construct the new range expression.
590  SourceRange ParenRange(Loop->getLParenLoc(), Loop->getRParenLoc());
591 
592  QualType Type = Context->getAutoDeductType();
593  if (!Descriptor.ElemType.isNull() && Descriptor.ElemType->isFundamentalType())
594  Type = Descriptor.ElemType.getUnqualifiedType();
595 
596  // If the new variable name is from the aliased variable, then the reference
597  // type for the new variable should only be used if the aliased variable was
598  // declared as a reference.
599  bool IsCheapToCopy =
600  !Descriptor.ElemType.isNull() &&
601  Descriptor.ElemType.isTriviallyCopyableType(*Context) &&
602  // TypeInfo::Width is in bits.
603  Context->getTypeInfo(Descriptor.ElemType).Width <= 8 * MaxCopySize;
604  bool UseCopy = CanCopy && ((VarNameFromAlias && !AliasVarIsRef) ||
605  (Descriptor.DerefByConstRef && IsCheapToCopy));
606 
607  if (!UseCopy) {
608  if (Descriptor.DerefByConstRef) {
609  Type = Context->getLValueReferenceType(Context->getConstType(Type));
610  } else if (Descriptor.DerefByValue) {
611  if (!IsCheapToCopy)
612  Type = Context->getRValueReferenceType(Type);
613  } else {
614  Type = Context->getLValueReferenceType(Type);
615  }
616  }
617 
618  StringRef MaybeDereference = Descriptor.ContainerNeedsDereference ? "*" : "";
619  std::string TypeString = Type.getAsString(getLangOpts());
620  std::string Range = ("(" + TypeString + " " + VarName + " : " +
621  MaybeDereference + Descriptor.ContainerString + ")")
622  .str();
623  Diag << FixItHint::CreateReplacement(
624  CharSourceRange::getTokenRange(ParenRange), Range);
625  TUInfo->getGeneratedDecls().insert(make_pair(Loop, VarName));
626 }
627 
628 /// \brief Returns a string which refers to the container iterated over.
629 StringRef LoopConvertCheck::getContainerString(ASTContext *Context,
630  const ForStmt *Loop,
631  const Expr *ContainerExpr) {
632  StringRef ContainerString;
633  if (isa<CXXThisExpr>(ContainerExpr->IgnoreParenImpCasts())) {
634  ContainerString = "this";
635  } else {
636  ContainerString =
637  getStringFromRange(Context->getSourceManager(), Context->getLangOpts(),
638  ContainerExpr->getSourceRange());
639  }
640 
641  return ContainerString;
642 }
643 
644 /// \brief Determines what kind of 'auto' must be used after converting a for
645 /// loop that iterates over an array or pseudoarray.
646 void LoopConvertCheck::getArrayLoopQualifiers(ASTContext *Context,
647  const BoundNodes &Nodes,
648  const Expr *ContainerExpr,
649  const UsageResult &Usages,
650  RangeDescriptor &Descriptor) {
651  // On arrays and pseudoarrays, we must figure out the qualifiers from the
652  // usages.
653  if (usagesAreConst(Context, Usages) ||
654  containerIsConst(ContainerExpr, Descriptor.ContainerNeedsDereference)) {
655  Descriptor.DerefByConstRef = true;
656  }
657  if (usagesReturnRValues(Usages)) {
658  // If the index usages (dereference, subscript, at, ...) return rvalues,
659  // then we should not use a reference, because we need to keep the code
660  // correct if it mutates the returned objects.
661  Descriptor.DerefByValue = true;
662  }
663  // Try to find the type of the elements on the container, to check if
664  // they are trivially copyable.
665  for (const Usage &U : Usages) {
666  if (!U.Expression || U.Expression->getType().isNull())
667  continue;
668  QualType Type = U.Expression->getType().getCanonicalType();
669  if (U.Kind == Usage::UK_MemberThroughArrow) {
670  if (!Type->isPointerType()) {
671  continue;
672  }
673  Type = Type->getPointeeType();
674  }
675  Descriptor.ElemType = Type;
676  }
677 }
678 
679 /// \brief Determines what kind of 'auto' must be used after converting an
680 /// iterator based for loop.
681 void LoopConvertCheck::getIteratorLoopQualifiers(ASTContext *Context,
682  const BoundNodes &Nodes,
683  RangeDescriptor &Descriptor) {
684  // The matchers for iterator loops provide bound nodes to obtain this
685  // information.
686  const auto *InitVar = Nodes.getDeclAs<VarDecl>(InitVarName);
687  QualType CanonicalInitVarType = InitVar->getType().getCanonicalType();
688  const auto *DerefByValueType =
689  Nodes.getNodeAs<QualType>(DerefByValueResultName);
690  Descriptor.DerefByValue = DerefByValueType;
691 
692  if (Descriptor.DerefByValue) {
693  // If the dereference operator returns by value then test for the
694  // canonical const qualification of the init variable type.
695  Descriptor.DerefByConstRef = CanonicalInitVarType.isConstQualified();
696  Descriptor.ElemType = *DerefByValueType;
697  } else {
698  if (const auto *DerefType =
699  Nodes.getNodeAs<QualType>(DerefByRefResultName)) {
700  // A node will only be bound with DerefByRefResultName if we're dealing
701  // with a user-defined iterator type. Test the const qualification of
702  // the reference type.
703  auto ValueType = DerefType->getNonReferenceType();
704 
705  Descriptor.DerefByConstRef = ValueType.isConstQualified();
706  Descriptor.ElemType = ValueType;
707  } else {
708  // By nature of the matcher this case is triggered only for built-in
709  // iterator types (i.e. pointers).
710  assert(isa<PointerType>(CanonicalInitVarType) &&
711  "Non-class iterator type is not a pointer type");
712 
713  // We test for const qualification of the pointed-at type.
714  Descriptor.DerefByConstRef =
715  CanonicalInitVarType->getPointeeType().isConstQualified();
716  Descriptor.ElemType = CanonicalInitVarType->getPointeeType();
717  }
718  }
719 }
720 
721 /// \brief Determines the parameters needed to build the range replacement.
722 void LoopConvertCheck::determineRangeDescriptor(
723  ASTContext *Context, const BoundNodes &Nodes, const ForStmt *Loop,
724  LoopFixerKind FixerKind, const Expr *ContainerExpr,
725  const UsageResult &Usages, RangeDescriptor &Descriptor) {
726  Descriptor.ContainerString = getContainerString(Context, Loop, ContainerExpr);
727 
728  if (FixerKind == LFK_Iterator)
729  getIteratorLoopQualifiers(Context, Nodes, Descriptor);
730  else
731  getArrayLoopQualifiers(Context, Nodes, ContainerExpr, Usages, Descriptor);
732 }
733 
734 /// \brief Check some of the conditions that must be met for the loop to be
735 /// convertible.
736 bool LoopConvertCheck::isConvertible(ASTContext *Context,
737  const ast_matchers::BoundNodes &Nodes,
738  const ForStmt *Loop,
739  LoopFixerKind FixerKind) {
740  // If we already modified the range of this for loop, don't do any further
741  // updates on this iteration.
742  if (TUInfo->getReplacedVars().count(Loop))
743  return false;
744 
745  // Check that we have exactly one index variable and at most one end variable.
746  const auto *LoopVar = Nodes.getDeclAs<VarDecl>(IncrementVarName);
747  const auto *CondVar = Nodes.getDeclAs<VarDecl>(ConditionVarName);
748  const auto *InitVar = Nodes.getDeclAs<VarDecl>(InitVarName);
749  if (!areSameVariable(LoopVar, CondVar) || !areSameVariable(LoopVar, InitVar))
750  return false;
751  const auto *EndVar = Nodes.getDeclAs<VarDecl>(EndVarName);
752  const auto *ConditionEndVar = Nodes.getDeclAs<VarDecl>(ConditionEndVarName);
753  if (EndVar && !areSameVariable(EndVar, ConditionEndVar))
754  return false;
755 
756  // FIXME: Try to put most of this logic inside a matcher.
757  if (FixerKind == LFK_Iterator) {
758  QualType InitVarType = InitVar->getType();
759  QualType CanonicalInitVarType = InitVarType.getCanonicalType();
760 
761  const auto *BeginCall = Nodes.getNodeAs<CXXMemberCallExpr>(BeginCallName);
762  assert(BeginCall && "Bad Callback. No begin call expression");
763  QualType CanonicalBeginType =
764  BeginCall->getMethodDecl()->getReturnType().getCanonicalType();
765  if (CanonicalBeginType->isPointerType() &&
766  CanonicalInitVarType->isPointerType()) {
767  // If the initializer and the variable are both pointers check if the
768  // un-qualified pointee types match, otherwise we don't use auto.
769  if (!Context->hasSameUnqualifiedType(
770  CanonicalBeginType->getPointeeType(),
771  CanonicalInitVarType->getPointeeType()))
772  return false;
773  } else if (!Context->hasSameType(CanonicalInitVarType,
774  CanonicalBeginType)) {
775  // Check for qualified types to avoid conversions from non-const to const
776  // iterator types.
777  return false;
778  }
779  } else if (FixerKind == LFK_PseudoArray) {
780  // This call is required to obtain the container.
781  const auto *EndCall = Nodes.getStmtAs<CXXMemberCallExpr>(EndCallName);
782  if (!EndCall || !dyn_cast<MemberExpr>(EndCall->getCallee()))
783  return false;
784  }
785  return true;
786 }
787 
788 void LoopConvertCheck::check(const MatchFinder::MatchResult &Result) {
789  const BoundNodes &Nodes = Result.Nodes;
790  Confidence ConfidenceLevel(Confidence::CL_Safe);
791  ASTContext *Context = Result.Context;
792 
793  const ForStmt *Loop;
794  LoopFixerKind FixerKind;
795  RangeDescriptor Descriptor;
796 
797  if ((Loop = Nodes.getStmtAs<ForStmt>(LoopNameArray))) {
798  FixerKind = LFK_Array;
799  } else if ((Loop = Nodes.getStmtAs<ForStmt>(LoopNameIterator))) {
800  FixerKind = LFK_Iterator;
801  } else {
802  Loop = Nodes.getStmtAs<ForStmt>(LoopNamePseudoArray);
803  assert(Loop && "Bad Callback. No for statement");
804  FixerKind = LFK_PseudoArray;
805  }
806 
807  if (!isConvertible(Context, Nodes, Loop, FixerKind))
808  return;
809 
810  const auto *LoopVar = Nodes.getDeclAs<VarDecl>(IncrementVarName);
811  const auto *EndVar = Nodes.getDeclAs<VarDecl>(EndVarName);
812 
813  // If the loop calls end()/size() after each iteration, lower our confidence
814  // level.
815  if (FixerKind != LFK_Array && !EndVar)
816  ConfidenceLevel.lowerTo(Confidence::CL_Reasonable);
817 
818  // If the end comparison isn't a variable, we can try to work with the
819  // expression the loop variable is being tested against instead.
820  const auto *EndCall = Nodes.getStmtAs<CXXMemberCallExpr>(EndCallName);
821  const auto *BoundExpr = Nodes.getStmtAs<Expr>(ConditionBoundName);
822 
823  // Find container expression of iterators and pseudoarrays, and determine if
824  // this expression needs to be dereferenced to obtain the container.
825  // With array loops, the container is often discovered during the
826  // ForLoopIndexUseVisitor traversal.
827  const Expr *ContainerExpr = nullptr;
828  if (FixerKind == LFK_Iterator) {
829  ContainerExpr = findContainer(Context, LoopVar->getInit(),
830  EndVar ? EndVar->getInit() : EndCall,
831  &Descriptor.ContainerNeedsDereference);
832  } else if (FixerKind == LFK_PseudoArray) {
833  ContainerExpr = EndCall->getImplicitObjectArgument();
834  Descriptor.ContainerNeedsDereference =
835  dyn_cast<MemberExpr>(EndCall->getCallee())->isArrow();
836  }
837 
838  // We must know the container or an array length bound.
839  if (!ContainerExpr && !BoundExpr)
840  return;
841 
842  ForLoopIndexUseVisitor Finder(Context, LoopVar, EndVar, ContainerExpr,
843  BoundExpr,
844  Descriptor.ContainerNeedsDereference);
845 
846  // Find expressions and variables on which the container depends.
847  if (ContainerExpr) {
848  ComponentFinderASTVisitor ComponentFinder;
849  ComponentFinder.findExprComponents(ContainerExpr->IgnoreParenImpCasts());
850  Finder.addComponents(ComponentFinder.getComponents());
851  }
852 
853  // Find usages of the loop index. If they are not used in a convertible way,
854  // stop here.
855  if (!Finder.findAndVerifyUsages(Loop->getBody()))
856  return;
857  ConfidenceLevel.lowerTo(Finder.getConfidenceLevel());
858 
859  // Obtain the container expression, if we don't have it yet.
860  if (FixerKind == LFK_Array) {
861  ContainerExpr = Finder.getContainerIndexed()->IgnoreParenImpCasts();
862 
863  // Very few loops are over expressions that generate arrays rather than
864  // array variables. Consider loops over arrays that aren't just represented
865  // by a variable to be risky conversions.
866  if (!getReferencedVariable(ContainerExpr) &&
867  !isDirectMemberExpr(ContainerExpr))
868  ConfidenceLevel.lowerTo(Confidence::CL_Risky);
869  }
870 
871  // Find out which qualifiers we have to use in the loop range.
872  const UsageResult &Usages = Finder.getUsages();
873  determineRangeDescriptor(Context, Nodes, Loop, FixerKind, ContainerExpr,
874  Usages, Descriptor);
875 
876  // Ensure that we do not try to move an expression dependent on a local
877  // variable declared inside the loop outside of it.
878  // FIXME: Determine when the external dependency isn't an expression converted
879  // by another loop.
880  TUInfo->getParentFinder().gatherAncestors(Context->getTranslationUnitDecl());
881  DependencyFinderASTVisitor DependencyFinder(
882  &TUInfo->getParentFinder().getStmtToParentStmtMap(),
883  &TUInfo->getParentFinder().getDeclToParentStmtMap(),
884  &TUInfo->getReplacedVars(), Loop);
885 
886  if (DependencyFinder.dependsOnInsideVariable(ContainerExpr) ||
887  Descriptor.ContainerString.empty() || Usages.empty() ||
888  ConfidenceLevel.getLevel() < MinConfidence)
889  return;
890 
891  doConversion(Context, LoopVar, getReferencedVariable(ContainerExpr), Usages,
892  Finder.getAliasDecl(), Finder.aliasUseRequired(),
893  Finder.aliasFromForInit(), Loop, Descriptor);
894 }
895 
896 } // namespace modernize
897 } // namespace tidy
898 } // namespace clang
static const char DerefByRefResultName[]
Discover usages of expressions consisting of index or iterator access.
const std::string Name
Definition: USRFinder.cpp:140
LangOptions LangOpts
Definition: ClangTidy.cpp:189
LangOptions getLangOpts() const
Returns the language options from the context.
Definition: ClangTidy.h:170
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
std::unique_ptr< ast_matchers::MatchFinder > Finder
Definition: ClangTidy.cpp:210
bool aliasFromForInit() const
Indicates if the alias declaration came from the init clause of a nested for loop.
static const char EndCallName[]
A class to encapsulate lowering of the tool's confidence level.
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:110
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...
const Expr * getContainerIndexed() const
Get the container indexed by IndexVar, if any.
static const char InitVarName[]
Level getLevel() const
Return the internal confidence level.
static const ValueDecl * getReferencedVariable(const Expr *E)
If the given expression is actually a DeclRefExpr or a MemberExpr, find and return the underlying Val...
SourceManager SourceMgr
Definition: ClangTidy.cpp:193
static const char EndVarName[]
Confidence::Level getConfidenceLevel() const
Accessor for ConfidenceLevel.
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't one.
void findExprComponents(const clang::Expr *SourceExpr)
Find the components of an expression and place them in a ComponentVector.
const ComponentVector & getComponents()
Accessor for Components.
SourceManager & SM
const DeclStmt * getAliasDecl() const
Returns the statement declaring the variable created as an alias for the loop element, if any.
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[]
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:385
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)...
const UsageResult & getUsages() const
Accessor for Usages.
static const char LoopNamePseudoArray[]
bool aliasUseRequired() const
Indicates if the alias declaration was in a place where it cannot simply be removed but rather replac...
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.
ClangTidyContext & Context
Definition: ClangTidy.cpp:93
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's name.
Definition: ClangTidy.cpp:352
const NamedDecl * Result
Definition: USRFinder.cpp:137