clang  8.0.0svn
TokenAnnotator.cpp
Go to the documentation of this file.
1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===//
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 /// \file
11 /// This file implements a token annotator, i.e. creates
12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information.
13 ///
14 //===----------------------------------------------------------------------===//
15 
16 #include "TokenAnnotator.h"
18 #include "llvm/ADT/SmallPtrSet.h"
19 #include "llvm/Support/Debug.h"
20 
21 #define DEBUG_TYPE "format-token-annotator"
22 
23 namespace clang {
24 namespace format {
25 
26 namespace {
27 
28 /// Returns \c true if the token can be used as an identifier in
29 /// an Objective-C \c @selector, \c false otherwise.
30 ///
31 /// Because getFormattingLangOpts() always lexes source code as
32 /// Objective-C++, C++ keywords like \c new and \c delete are
33 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C.
34 ///
35 /// For Objective-C and Objective-C++, both identifiers and keywords
36 /// are valid inside @selector(...) (or a macro which
37 /// invokes @selector(...)). So, we allow treat any identifier or
38 /// keyword as a potential Objective-C selector component.
39 static bool canBeObjCSelectorComponent(const FormatToken &Tok) {
40  return Tok.Tok.getIdentifierInfo() != nullptr;
41 }
42 
43 /// A parser that gathers additional information about tokens.
44 ///
45 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
46 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
47 /// into template parameter lists.
48 class AnnotatingParser {
49 public:
50  AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line,
51  const AdditionalKeywords &Keywords)
52  : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false),
53  Keywords(Keywords) {
54  Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
55  resetTokenMetadata(CurrentToken);
56  }
57 
58 private:
59  bool parseAngle() {
60  if (!CurrentToken || !CurrentToken->Previous)
61  return false;
62  if (NonTemplateLess.count(CurrentToken->Previous))
63  return false;
64 
65  const FormatToken &Previous = *CurrentToken->Previous; // The '<'.
66  if (Previous.Previous) {
67  if (Previous.Previous->Tok.isLiteral())
68  return false;
69  if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 &&
70  (!Previous.Previous->MatchingParen ||
71  !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen)))
72  return false;
73  }
74 
75  FormatToken *Left = CurrentToken->Previous;
76  Left->ParentBracket = Contexts.back().ContextKind;
77  ScopedContextCreator ContextCreator(*this, tok::less, 12);
78 
79  // If this angle is in the context of an expression, we need to be more
80  // hesitant to detect it as opening template parameters.
81  bool InExprContext = Contexts.back().IsExpression;
82 
83  Contexts.back().IsExpression = false;
84  // If there's a template keyword before the opening angle bracket, this is a
85  // template parameter, not an argument.
86  Contexts.back().InTemplateArgument =
87  Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
88 
89  if (Style.Language == FormatStyle::LK_Java &&
90  CurrentToken->is(tok::question))
91  next();
92 
93  while (CurrentToken) {
94  if (CurrentToken->is(tok::greater)) {
95  Left->MatchingParen = CurrentToken;
96  CurrentToken->MatchingParen = Left;
97  // In TT_Proto, we must distignuish between:
98  // map<key, value>
99  // msg < item: data >
100  // msg: < item: data >
101  // In TT_TextProto, map<key, value> does not occur.
102  if (Style.Language == FormatStyle::LK_TextProto ||
103  (Style.Language == FormatStyle::LK_Proto && Left->Previous &&
104  Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral)))
105  CurrentToken->Type = TT_DictLiteral;
106  else
107  CurrentToken->Type = TT_TemplateCloser;
108  next();
109  return true;
110  }
111  if (CurrentToken->is(tok::question) &&
112  Style.Language == FormatStyle::LK_Java) {
113  next();
114  continue;
115  }
116  if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
117  (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext &&
118  Style.Language != FormatStyle::LK_Proto &&
119  Style.Language != FormatStyle::LK_TextProto))
120  return false;
121  // If a && or || is found and interpreted as a binary operator, this set
122  // of angles is likely part of something like "a < b && c > d". If the
123  // angles are inside an expression, the ||/&& might also be a binary
124  // operator that was misinterpreted because we are parsing template
125  // parameters.
126  // FIXME: This is getting out of hand, write a decent parser.
127  if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
128  CurrentToken->Previous->is(TT_BinaryOperator) &&
129  Contexts[Contexts.size() - 2].IsExpression &&
130  !Line.startsWith(tok::kw_template))
131  return false;
132  updateParameterCount(Left, CurrentToken);
133  if (Style.Language == FormatStyle::LK_Proto) {
134  if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) {
135  if (CurrentToken->is(tok::colon) ||
136  (CurrentToken->isOneOf(tok::l_brace, tok::less) &&
137  Previous->isNot(tok::colon)))
138  Previous->Type = TT_SelectorName;
139  }
140  }
141  if (!consumeToken())
142  return false;
143  }
144  return false;
145  }
146 
147  bool parseParens(bool LookForDecls = false) {
148  if (!CurrentToken)
149  return false;
150  FormatToken *Left = CurrentToken->Previous;
151  Left->ParentBracket = Contexts.back().ContextKind;
152  ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
153 
154  // FIXME: This is a bit of a hack. Do better.
155  Contexts.back().ColonIsForRangeExpr =
156  Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
157 
158  bool StartsObjCMethodExpr = false;
159  if (FormatToken *MaybeSel = Left->Previous) {
160  // @selector( starts a selector.
161  if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
162  MaybeSel->Previous->is(tok::at)) {
163  StartsObjCMethodExpr = true;
164  }
165  }
166 
167  if (Left->is(TT_OverloadedOperatorLParen)) {
168  Contexts.back().IsExpression = false;
169  } else if (Style.Language == FormatStyle::LK_JavaScript &&
170  (Line.startsWith(Keywords.kw_type, tok::identifier) ||
171  Line.startsWith(tok::kw_export, Keywords.kw_type,
172  tok::identifier))) {
173  // type X = (...);
174  // export type X = (...);
175  Contexts.back().IsExpression = false;
176  } else if (Left->Previous &&
177  (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype,
178  tok::kw_if, tok::kw_while, tok::l_paren,
179  tok::comma) ||
180  Left->Previous->endsSequence(tok::kw_constexpr, tok::kw_if) ||
181  Left->Previous->is(TT_BinaryOperator))) {
182  // static_assert, if and while usually contain expressions.
183  Contexts.back().IsExpression = true;
184  } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
185  (Left->Previous->is(Keywords.kw_function) ||
186  (Left->Previous->endsSequence(tok::identifier,
187  Keywords.kw_function)))) {
188  // function(...) or function f(...)
189  Contexts.back().IsExpression = false;
190  } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
191  Left->Previous->is(TT_JsTypeColon)) {
192  // let x: (SomeType);
193  Contexts.back().IsExpression = false;
194  } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
195  Left->Previous->MatchingParen &&
196  Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
197  // This is a parameter list of a lambda expression.
198  Contexts.back().IsExpression = false;
199  } else if (Line.InPPDirective &&
200  (!Left->Previous || !Left->Previous->is(tok::identifier))) {
201  Contexts.back().IsExpression = true;
202  } else if (Contexts[Contexts.size() - 2].CaretFound) {
203  // This is the parameter list of an ObjC block.
204  Contexts.back().IsExpression = false;
205  } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
206  Left->Type = TT_AttributeParen;
207  } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
208  // The first argument to a foreach macro is a declaration.
209  Contexts.back().IsForEachMacro = true;
210  Contexts.back().IsExpression = false;
211  } else if (Left->Previous && Left->Previous->MatchingParen &&
212  Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
213  Contexts.back().IsExpression = false;
214  } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
215  bool IsForOrCatch =
216  Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
217  Contexts.back().IsExpression = !IsForOrCatch;
218  }
219 
220  if (StartsObjCMethodExpr) {
221  Contexts.back().ColonIsObjCMethodExpr = true;
222  Left->Type = TT_ObjCMethodExpr;
223  }
224 
225  // MightBeFunctionType and ProbablyFunctionType are used for
226  // function pointer and reference types as well as Objective-C
227  // block types:
228  //
229  // void (*FunctionPointer)(void);
230  // void (&FunctionReference)(void);
231  // void (^ObjCBlock)(void);
232  bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression;
233  bool ProbablyFunctionType =
234  CurrentToken->isOneOf(tok::star, tok::amp, tok::caret);
235  bool HasMultipleLines = false;
236  bool HasMultipleParametersOnALine = false;
237  bool MightBeObjCForRangeLoop =
238  Left->Previous && Left->Previous->is(tok::kw_for);
239  FormatToken *PossibleObjCForInToken = nullptr;
240  while (CurrentToken) {
241  // LookForDecls is set when "if (" has been seen. Check for
242  // 'identifier' '*' 'identifier' followed by not '=' -- this
243  // '*' has to be a binary operator but determineStarAmpUsage() will
244  // categorize it as an unary operator, so set the right type here.
245  if (LookForDecls && CurrentToken->Next) {
246  FormatToken *Prev = CurrentToken->getPreviousNonComment();
247  if (Prev) {
248  FormatToken *PrevPrev = Prev->getPreviousNonComment();
249  FormatToken *Next = CurrentToken->Next;
250  if (PrevPrev && PrevPrev->is(tok::identifier) &&
251  Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
252  CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
253  Prev->Type = TT_BinaryOperator;
254  LookForDecls = false;
255  }
256  }
257  }
258 
259  if (CurrentToken->Previous->is(TT_PointerOrReference) &&
260  CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
261  tok::coloncolon))
262  ProbablyFunctionType = true;
263  if (CurrentToken->is(tok::comma))
264  MightBeFunctionType = false;
265  if (CurrentToken->Previous->is(TT_BinaryOperator))
266  Contexts.back().IsExpression = true;
267  if (CurrentToken->is(tok::r_paren)) {
268  if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next &&
269  (CurrentToken->Next->is(tok::l_paren) ||
270  (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration)))
271  Left->Type = Left->Next->is(tok::caret) ? TT_ObjCBlockLParen
272  : TT_FunctionTypeLParen;
273  Left->MatchingParen = CurrentToken;
274  CurrentToken->MatchingParen = Left;
275 
276  if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) &&
277  Left->Previous && Left->Previous->is(tok::l_paren)) {
278  // Detect the case where macros are used to generate lambdas or
279  // function bodies, e.g.:
280  // auto my_lambda = MARCO((Type *type, int i) { .. body .. });
281  for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) {
282  if (Tok->is(TT_BinaryOperator) &&
283  Tok->isOneOf(tok::star, tok::amp, tok::ampamp))
284  Tok->Type = TT_PointerOrReference;
285  }
286  }
287 
288  if (StartsObjCMethodExpr) {
289  CurrentToken->Type = TT_ObjCMethodExpr;
290  if (Contexts.back().FirstObjCSelectorName) {
291  Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
292  Contexts.back().LongestObjCSelectorName;
293  }
294  }
295 
296  if (Left->is(TT_AttributeParen))
297  CurrentToken->Type = TT_AttributeParen;
298  if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
299  CurrentToken->Type = TT_JavaAnnotation;
300  if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
301  CurrentToken->Type = TT_LeadingJavaAnnotation;
302 
303  if (!HasMultipleLines)
304  Left->PackingKind = PPK_Inconclusive;
305  else if (HasMultipleParametersOnALine)
306  Left->PackingKind = PPK_BinPacked;
307  else
308  Left->PackingKind = PPK_OnePerLine;
309 
310  next();
311  return true;
312  }
313  if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
314  return false;
315 
316  if (CurrentToken->is(tok::l_brace))
317  Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
318  if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
319  !CurrentToken->Next->HasUnescapedNewline &&
320  !CurrentToken->Next->isTrailingComment())
321  HasMultipleParametersOnALine = true;
322  if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) ||
323  CurrentToken->Previous->isSimpleTypeSpecifier()) &&
324  !CurrentToken->is(tok::l_brace))
325  Contexts.back().IsExpression = false;
326  if (CurrentToken->isOneOf(tok::semi, tok::colon)) {
327  MightBeObjCForRangeLoop = false;
328  if (PossibleObjCForInToken) {
329  PossibleObjCForInToken->Type = TT_Unknown;
330  PossibleObjCForInToken = nullptr;
331  }
332  }
333  if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) {
334  PossibleObjCForInToken = CurrentToken;
335  PossibleObjCForInToken->Type = TT_ObjCForIn;
336  }
337  // When we discover a 'new', we set CanBeExpression to 'false' in order to
338  // parse the type correctly. Reset that after a comma.
339  if (CurrentToken->is(tok::comma))
340  Contexts.back().CanBeExpression = true;
341 
342  FormatToken *Tok = CurrentToken;
343  if (!consumeToken())
344  return false;
345  updateParameterCount(Left, Tok);
346  if (CurrentToken && CurrentToken->HasUnescapedNewline)
347  HasMultipleLines = true;
348  }
349  return false;
350  }
351 
352  bool isCpp11AttributeSpecifier(const FormatToken &Tok) {
353  if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square))
354  return false;
355  const FormatToken *AttrTok = Tok.Next->Next;
356  if (!AttrTok)
357  return false;
358  // C++17 '[[using ns: foo, bar(baz, blech)]]'
359  // We assume nobody will name an ObjC variable 'using'.
360  if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon))
361  return true;
362  if (AttrTok->isNot(tok::identifier))
363  return false;
364  while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) {
365  // ObjC message send. We assume nobody will use : in a C++11 attribute
366  // specifier parameter, although this is technically valid:
367  // [[foo(:)]]
368  if (AttrTok->is(tok::colon) ||
369  AttrTok->startsSequence(tok::identifier, tok::identifier) ||
370  AttrTok->startsSequence(tok::r_paren, tok::identifier))
371  return false;
372  if (AttrTok->is(tok::ellipsis))
373  return true;
374  AttrTok = AttrTok->Next;
375  }
376  return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square);
377  }
378 
379  bool parseSquare() {
380  if (!CurrentToken)
381  return false;
382 
383  // A '[' could be an index subscript (after an identifier or after
384  // ')' or ']'), it could be the start of an Objective-C method
385  // expression, it could the start of an Objective-C array literal,
386  // or it could be a C++ attribute specifier [[foo::bar]].
387  FormatToken *Left = CurrentToken->Previous;
388  Left->ParentBracket = Contexts.back().ContextKind;
389  FormatToken *Parent = Left->getPreviousNonComment();
390 
391  // Cases where '>' is followed by '['.
392  // In C++, this can happen either in array of templates (foo<int>[10])
393  // or when array is a nested template type (unique_ptr<type1<type2>[]>).
394  bool CppArrayTemplates =
395  Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) &&
396  (Contexts.back().CanBeExpression || Contexts.back().IsExpression ||
397  Contexts.back().InTemplateArgument);
398 
399  bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) ||
400  Contexts.back().InCpp11AttributeSpecifier;
401 
402  bool InsideInlineASM = Line.startsWith(tok::kw_asm);
403  bool StartsObjCMethodExpr =
404  !InsideInlineASM && !CppArrayTemplates && Style.isCpp() &&
405  !IsCpp11AttributeSpecifier && Contexts.back().CanBeExpression &&
406  Left->isNot(TT_LambdaLSquare) &&
407  !CurrentToken->isOneOf(tok::l_brace, tok::r_square) &&
408  (!Parent ||
409  Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
410  tok::kw_return, tok::kw_throw) ||
411  Parent->isUnaryOperator() ||
412  // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
413  Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
414  getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
415  bool ColonFound = false;
416 
417  unsigned BindingIncrease = 1;
418  if (Left->isCppStructuredBinding(Style)) {
419  Left->Type = TT_StructuredBindingLSquare;
420  } else if (Left->is(TT_Unknown)) {
421  if (StartsObjCMethodExpr) {
422  Left->Type = TT_ObjCMethodExpr;
423  } else if (IsCpp11AttributeSpecifier) {
424  Left->Type = TT_AttributeSquare;
425  } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
426  Contexts.back().ContextKind == tok::l_brace &&
427  Parent->isOneOf(tok::l_brace, tok::comma)) {
428  Left->Type = TT_JsComputedPropertyName;
429  } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace &&
430  Parent && Parent->isOneOf(tok::l_brace, tok::comma)) {
431  Left->Type = TT_DesignatedInitializerLSquare;
432  } else if (CurrentToken->is(tok::r_square) && Parent &&
433  Parent->is(TT_TemplateCloser)) {
434  Left->Type = TT_ArraySubscriptLSquare;
435  } else if (Style.Language == FormatStyle::LK_Proto ||
436  Style.Language == FormatStyle::LK_TextProto) {
437  // Square braces in LK_Proto can either be message field attributes:
438  //
439  // optional Aaa aaa = 1 [
440  // (aaa) = aaa
441  // ];
442  //
443  // extensions 123 [
444  // (aaa) = aaa
445  // ];
446  //
447  // or text proto extensions (in options):
448  //
449  // option (Aaa.options) = {
450  // [type.type/type] {
451  // key: value
452  // }
453  // }
454  //
455  // or repeated fields (in options):
456  //
457  // option (Aaa.options) = {
458  // keys: [ 1, 2, 3 ]
459  // }
460  //
461  // In the first and the third case we want to spread the contents inside
462  // the square braces; in the second we want to keep them inline.
463  Left->Type = TT_ArrayInitializerLSquare;
464  if (!Left->endsSequence(tok::l_square, tok::numeric_constant,
465  tok::equal) &&
466  !Left->endsSequence(tok::l_square, tok::numeric_constant,
467  tok::identifier) &&
468  !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) {
469  Left->Type = TT_ProtoExtensionLSquare;
470  BindingIncrease = 10;
471  }
472  } else if (!CppArrayTemplates && Parent &&
473  Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at,
474  tok::comma, tok::l_paren, tok::l_square,
475  tok::question, tok::colon, tok::kw_return,
476  // Should only be relevant to JavaScript:
477  tok::kw_default)) {
478  Left->Type = TT_ArrayInitializerLSquare;
479  } else {
480  BindingIncrease = 10;
481  Left->Type = TT_ArraySubscriptLSquare;
482  }
483  }
484 
485  ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
486  Contexts.back().IsExpression = true;
487  if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
488  Parent->is(TT_JsTypeColon))
489  Contexts.back().IsExpression = false;
490 
491  Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
492  Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier;
493 
494  while (CurrentToken) {
495  if (CurrentToken->is(tok::r_square)) {
496  if (IsCpp11AttributeSpecifier)
497  CurrentToken->Type = TT_AttributeSquare;
498  else if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
499  Left->is(TT_ObjCMethodExpr)) {
500  // An ObjC method call is rarely followed by an open parenthesis.
501  // FIXME: Do we incorrectly label ":" with this?
502  StartsObjCMethodExpr = false;
503  Left->Type = TT_Unknown;
504  }
505  if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
506  CurrentToken->Type = TT_ObjCMethodExpr;
507  // If we haven't seen a colon yet, make sure the last identifier
508  // before the r_square is tagged as a selector name component.
509  if (!ColonFound && CurrentToken->Previous &&
510  CurrentToken->Previous->is(TT_Unknown) &&
511  canBeObjCSelectorComponent(*CurrentToken->Previous))
512  CurrentToken->Previous->Type = TT_SelectorName;
513  // determineStarAmpUsage() thinks that '*' '[' is allocating an
514  // array of pointers, but if '[' starts a selector then '*' is a
515  // binary operator.
516  if (Parent && Parent->is(TT_PointerOrReference))
517  Parent->Type = TT_BinaryOperator;
518  }
519  Left->MatchingParen = CurrentToken;
520  CurrentToken->MatchingParen = Left;
521  // FirstObjCSelectorName is set when a colon is found. This does
522  // not work, however, when the method has no parameters.
523  // Here, we set FirstObjCSelectorName when the end of the method call is
524  // reached, in case it was not set already.
525  if (!Contexts.back().FirstObjCSelectorName) {
526  FormatToken* Previous = CurrentToken->getPreviousNonComment();
527  if (Previous && Previous->is(TT_SelectorName)) {
528  Previous->ObjCSelectorNameParts = 1;
529  Contexts.back().FirstObjCSelectorName = Previous;
530  }
531  } else {
532  Left->ParameterCount =
533  Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
534  }
535  if (Contexts.back().FirstObjCSelectorName) {
536  Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
537  Contexts.back().LongestObjCSelectorName;
538  if (Left->BlockParameterCount > 1)
539  Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
540  }
541  next();
542  return true;
543  }
544  if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
545  return false;
546  if (CurrentToken->is(tok::colon)) {
547  if (IsCpp11AttributeSpecifier &&
548  CurrentToken->endsSequence(tok::colon, tok::identifier,
549  tok::kw_using)) {
550  // Remember that this is a [[using ns: foo]] C++ attribute, so we
551  // don't add a space before the colon (unlike other colons).
552  CurrentToken->Type = TT_AttributeColon;
553  } else if (Left->isOneOf(TT_ArraySubscriptLSquare,
554  TT_DesignatedInitializerLSquare)) {
555  Left->Type = TT_ObjCMethodExpr;
556  StartsObjCMethodExpr = true;
557  Contexts.back().ColonIsObjCMethodExpr = true;
558  if (Parent && Parent->is(tok::r_paren))
559  // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
560  Parent->Type = TT_CastRParen;
561  }
562  ColonFound = true;
563  }
564  if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
565  !ColonFound)
566  Left->Type = TT_ArrayInitializerLSquare;
567  FormatToken *Tok = CurrentToken;
568  if (!consumeToken())
569  return false;
570  updateParameterCount(Left, Tok);
571  }
572  return false;
573  }
574 
575  bool parseBrace() {
576  if (CurrentToken) {
577  FormatToken *Left = CurrentToken->Previous;
578  Left->ParentBracket = Contexts.back().ContextKind;
579 
580  if (Contexts.back().CaretFound)
581  Left->Type = TT_ObjCBlockLBrace;
582  Contexts.back().CaretFound = false;
583 
584  ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
585  Contexts.back().ColonIsDictLiteral = true;
586  if (Left->BlockKind == BK_BracedInit)
587  Contexts.back().IsExpression = true;
588  if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
589  Left->Previous->is(TT_JsTypeColon))
590  Contexts.back().IsExpression = false;
591 
592  while (CurrentToken) {
593  if (CurrentToken->is(tok::r_brace)) {
594  Left->MatchingParen = CurrentToken;
595  CurrentToken->MatchingParen = Left;
596  next();
597  return true;
598  }
599  if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
600  return false;
601  updateParameterCount(Left, CurrentToken);
602  if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) {
603  FormatToken *Previous = CurrentToken->getPreviousNonComment();
604  if (Previous->is(TT_JsTypeOptionalQuestion))
605  Previous = Previous->getPreviousNonComment();
606  if ((CurrentToken->is(tok::colon) &&
607  (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) ||
608  Style.Language == FormatStyle::LK_Proto ||
609  Style.Language == FormatStyle::LK_TextProto) {
610  Left->Type = TT_DictLiteral;
611  if (Previous->Tok.getIdentifierInfo() ||
612  Previous->is(tok::string_literal))
613  Previous->Type = TT_SelectorName;
614  }
615  if (CurrentToken->is(tok::colon) ||
616  Style.Language == FormatStyle::LK_JavaScript)
617  Left->Type = TT_DictLiteral;
618  }
619  if (CurrentToken->is(tok::comma) &&
620  Style.Language == FormatStyle::LK_JavaScript)
621  Left->Type = TT_DictLiteral;
622  if (!consumeToken())
623  return false;
624  }
625  }
626  return true;
627  }
628 
629  void updateParameterCount(FormatToken *Left, FormatToken *Current) {
630  // For ObjC methods, the number of parameters is calculated differently as
631  // method declarations have a different structure (the parameters are not
632  // inside a bracket scope).
633  if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block)
634  ++Left->BlockParameterCount;
635  if (Current->is(tok::comma)) {
636  ++Left->ParameterCount;
637  if (!Left->Role)
638  Left->Role.reset(new CommaSeparatedList(Style));
639  Left->Role->CommaFound(Current);
640  } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
641  Left->ParameterCount = 1;
642  }
643  }
644 
645  bool parseConditional() {
646  while (CurrentToken) {
647  if (CurrentToken->is(tok::colon)) {
648  CurrentToken->Type = TT_ConditionalExpr;
649  next();
650  return true;
651  }
652  if (!consumeToken())
653  return false;
654  }
655  return false;
656  }
657 
658  bool parseTemplateDeclaration() {
659  if (CurrentToken && CurrentToken->is(tok::less)) {
660  CurrentToken->Type = TT_TemplateOpener;
661  next();
662  if (!parseAngle())
663  return false;
664  if (CurrentToken)
665  CurrentToken->Previous->ClosesTemplateDeclaration = true;
666  return true;
667  }
668  return false;
669  }
670 
671  bool consumeToken() {
672  FormatToken *Tok = CurrentToken;
673  next();
674  switch (Tok->Tok.getKind()) {
675  case tok::plus:
676  case tok::minus:
677  if (!Tok->Previous && Line.MustBeDeclaration)
678  Tok->Type = TT_ObjCMethodSpecifier;
679  break;
680  case tok::colon:
681  if (!Tok->Previous)
682  return false;
683  // Colons from ?: are handled in parseConditional().
684  if (Style.Language == FormatStyle::LK_JavaScript) {
685  if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
686  (Contexts.size() == 1 && // switch/case labels
687  !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
688  Contexts.back().ContextKind == tok::l_paren || // function params
689  Contexts.back().ContextKind == tok::l_square || // array type
690  (!Contexts.back().IsExpression &&
691  Contexts.back().ContextKind == tok::l_brace) || // object type
692  (Contexts.size() == 1 &&
693  Line.MustBeDeclaration)) { // method/property declaration
694  Contexts.back().IsExpression = false;
695  Tok->Type = TT_JsTypeColon;
696  break;
697  }
698  }
699  if (Contexts.back().ColonIsDictLiteral ||
700  Style.Language == FormatStyle::LK_Proto ||
701  Style.Language == FormatStyle::LK_TextProto) {
702  Tok->Type = TT_DictLiteral;
703  if (Style.Language == FormatStyle::LK_TextProto) {
704  if (FormatToken *Previous = Tok->getPreviousNonComment())
705  Previous->Type = TT_SelectorName;
706  }
707  } else if (Contexts.back().ColonIsObjCMethodExpr ||
708  Line.startsWith(TT_ObjCMethodSpecifier)) {
709  Tok->Type = TT_ObjCMethodExpr;
710  const FormatToken *BeforePrevious = Tok->Previous->Previous;
711  // Ensure we tag all identifiers in method declarations as
712  // TT_SelectorName.
713  bool UnknownIdentifierInMethodDeclaration =
714  Line.startsWith(TT_ObjCMethodSpecifier) &&
715  Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown);
716  if (!BeforePrevious ||
717  // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
718  !(BeforePrevious->is(TT_CastRParen) ||
719  (BeforePrevious->is(TT_ObjCMethodExpr) &&
720  BeforePrevious->is(tok::colon))) ||
721  BeforePrevious->is(tok::r_square) ||
722  Contexts.back().LongestObjCSelectorName == 0 ||
723  UnknownIdentifierInMethodDeclaration) {
724  Tok->Previous->Type = TT_SelectorName;
725  if (!Contexts.back().FirstObjCSelectorName)
726  Contexts.back().FirstObjCSelectorName = Tok->Previous;
727  else if (Tok->Previous->ColumnWidth >
728  Contexts.back().LongestObjCSelectorName)
729  Contexts.back().LongestObjCSelectorName =
730  Tok->Previous->ColumnWidth;
731  Tok->Previous->ParameterIndex =
732  Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
733  ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
734  }
735  } else if (Contexts.back().ColonIsForRangeExpr) {
736  Tok->Type = TT_RangeBasedForLoopColon;
737  } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
738  Tok->Type = TT_BitFieldColon;
739  } else if (Contexts.size() == 1 &&
740  !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
741  if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren,
742  tok::kw_noexcept))
743  Tok->Type = TT_CtorInitializerColon;
744  else
745  Tok->Type = TT_InheritanceColon;
746  } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next &&
747  (Tok->Next->isOneOf(tok::r_paren, tok::comma) ||
748  (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next &&
749  Tok->Next->Next->is(tok::colon)))) {
750  // This handles a special macro in ObjC code where selectors including
751  // the colon are passed as macro arguments.
752  Tok->Type = TT_ObjCMethodExpr;
753  } else if (Contexts.back().ContextKind == tok::l_paren) {
754  Tok->Type = TT_InlineASMColon;
755  }
756  break;
757  case tok::pipe:
758  case tok::amp:
759  // | and & in declarations/type expressions represent union and
760  // intersection types, respectively.
761  if (Style.Language == FormatStyle::LK_JavaScript &&
762  !Contexts.back().IsExpression)
763  Tok->Type = TT_JsTypeOperator;
764  break;
765  case tok::kw_if:
766  case tok::kw_while:
767  if (Tok->is(tok::kw_if) && CurrentToken &&
768  CurrentToken->is(tok::kw_constexpr))
769  next();
770  if (CurrentToken && CurrentToken->is(tok::l_paren)) {
771  next();
772  if (!parseParens(/*LookForDecls=*/true))
773  return false;
774  }
775  break;
776  case tok::kw_for:
777  if (Style.Language == FormatStyle::LK_JavaScript) {
778  // x.for and {for: ...}
779  if ((Tok->Previous && Tok->Previous->is(tok::period)) ||
780  (Tok->Next && Tok->Next->is(tok::colon)))
781  break;
782  // JS' for await ( ...
783  if (CurrentToken && CurrentToken->is(Keywords.kw_await))
784  next();
785  }
786  Contexts.back().ColonIsForRangeExpr = true;
787  next();
788  if (!parseParens())
789  return false;
790  break;
791  case tok::l_paren:
792  // When faced with 'operator()()', the kw_operator handler incorrectly
793  // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
794  // the first two parens OverloadedOperators and the second l_paren an
795  // OverloadedOperatorLParen.
796  if (Tok->Previous && Tok->Previous->is(tok::r_paren) &&
797  Tok->Previous->MatchingParen &&
798  Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
799  Tok->Previous->Type = TT_OverloadedOperator;
800  Tok->Previous->MatchingParen->Type = TT_OverloadedOperator;
801  Tok->Type = TT_OverloadedOperatorLParen;
802  }
803 
804  if (!parseParens())
805  return false;
806  if (Line.MustBeDeclaration && Contexts.size() == 1 &&
807  !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
808  (!Tok->Previous ||
809  !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute,
810  TT_LeadingJavaAnnotation)))
811  Line.MightBeFunctionDecl = true;
812  break;
813  case tok::l_square:
814  if (!parseSquare())
815  return false;
816  break;
817  case tok::l_brace:
818  if (Style.Language == FormatStyle::LK_TextProto) {
819  FormatToken *Previous = Tok->getPreviousNonComment();
820  if (Previous && Previous->Type != TT_DictLiteral)
821  Previous->Type = TT_SelectorName;
822  }
823  if (!parseBrace())
824  return false;
825  break;
826  case tok::less:
827  if (parseAngle()) {
828  Tok->Type = TT_TemplateOpener;
829  // In TT_Proto, we must distignuish between:
830  // map<key, value>
831  // msg < item: data >
832  // msg: < item: data >
833  // In TT_TextProto, map<key, value> does not occur.
834  if (Style.Language == FormatStyle::LK_TextProto ||
835  (Style.Language == FormatStyle::LK_Proto && Tok->Previous &&
836  Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) {
837  Tok->Type = TT_DictLiteral;
838  FormatToken *Previous = Tok->getPreviousNonComment();
839  if (Previous && Previous->Type != TT_DictLiteral)
840  Previous->Type = TT_SelectorName;
841  }
842  } else {
843  Tok->Type = TT_BinaryOperator;
844  NonTemplateLess.insert(Tok);
845  CurrentToken = Tok;
846  next();
847  }
848  break;
849  case tok::r_paren:
850  case tok::r_square:
851  return false;
852  case tok::r_brace:
853  // Lines can start with '}'.
854  if (Tok->Previous)
855  return false;
856  break;
857  case tok::greater:
858  if (Style.Language != FormatStyle::LK_TextProto)
859  Tok->Type = TT_BinaryOperator;
860  break;
861  case tok::kw_operator:
862  if (Style.Language == FormatStyle::LK_TextProto ||
863  Style.Language == FormatStyle::LK_Proto)
864  break;
865  while (CurrentToken &&
866  !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
867  if (CurrentToken->isOneOf(tok::star, tok::amp))
868  CurrentToken->Type = TT_PointerOrReference;
869  consumeToken();
870  if (CurrentToken &&
871  CurrentToken->Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator,
872  tok::comma))
873  CurrentToken->Previous->Type = TT_OverloadedOperator;
874  }
875  if (CurrentToken) {
876  CurrentToken->Type = TT_OverloadedOperatorLParen;
877  if (CurrentToken->Previous->is(TT_BinaryOperator))
878  CurrentToken->Previous->Type = TT_OverloadedOperator;
879  }
880  break;
881  case tok::question:
882  if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
883  Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
884  tok::r_brace)) {
885  // Question marks before semicolons, colons, etc. indicate optional
886  // types (fields, parameters), e.g.
887  // function(x?: string, y?) {...}
888  // class X { y?; }
889  Tok->Type = TT_JsTypeOptionalQuestion;
890  break;
891  }
892  // Declarations cannot be conditional expressions, this can only be part
893  // of a type declaration.
894  if (Line.MustBeDeclaration && !Contexts.back().IsExpression &&
895  Style.Language == FormatStyle::LK_JavaScript)
896  break;
897  parseConditional();
898  break;
899  case tok::kw_template:
900  parseTemplateDeclaration();
901  break;
902  case tok::comma:
903  if (Contexts.back().InCtorInitializer)
904  Tok->Type = TT_CtorInitializerComma;
905  else if (Contexts.back().InInheritanceList)
906  Tok->Type = TT_InheritanceComma;
907  else if (Contexts.back().FirstStartOfName &&
908  (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
909  Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
910  Line.IsMultiVariableDeclStmt = true;
911  }
912  if (Contexts.back().IsForEachMacro)
913  Contexts.back().IsExpression = true;
914  break;
915  case tok::identifier:
916  if (Tok->isOneOf(Keywords.kw___has_include,
917  Keywords.kw___has_include_next)) {
918  parseHasInclude();
919  }
920  break;
921  default:
922  break;
923  }
924  return true;
925  }
926 
927  void parseIncludeDirective() {
928  if (CurrentToken && CurrentToken->is(tok::less)) {
929  next();
930  while (CurrentToken) {
931  // Mark tokens up to the trailing line comments as implicit string
932  // literals.
933  if (CurrentToken->isNot(tok::comment) &&
934  !CurrentToken->TokenText.startswith("//"))
935  CurrentToken->Type = TT_ImplicitStringLiteral;
936  next();
937  }
938  }
939  }
940 
941  void parseWarningOrError() {
942  next();
943  // We still want to format the whitespace left of the first token of the
944  // warning or error.
945  next();
946  while (CurrentToken) {
947  CurrentToken->Type = TT_ImplicitStringLiteral;
948  next();
949  }
950  }
951 
952  void parsePragma() {
953  next(); // Consume "pragma".
954  if (CurrentToken &&
955  CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
956  bool IsMark = CurrentToken->is(Keywords.kw_mark);
957  next(); // Consume "mark".
958  next(); // Consume first token (so we fix leading whitespace).
959  while (CurrentToken) {
960  if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
961  CurrentToken->Type = TT_ImplicitStringLiteral;
962  next();
963  }
964  }
965  }
966 
967  void parseHasInclude() {
968  if (!CurrentToken || !CurrentToken->is(tok::l_paren))
969  return;
970  next(); // '('
971  parseIncludeDirective();
972  next(); // ')'
973  }
974 
975  LineType parsePreprocessorDirective() {
976  bool IsFirstToken = CurrentToken->IsFirst;
978  next();
979  if (!CurrentToken)
980  return Type;
981 
982  if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) {
983  // JavaScript files can contain shebang lines of the form:
984  // #!/usr/bin/env node
985  // Treat these like C++ #include directives.
986  while (CurrentToken) {
987  // Tokens cannot be comments here.
988  CurrentToken->Type = TT_ImplicitStringLiteral;
989  next();
990  }
991  return LT_ImportStatement;
992  }
993 
994  if (CurrentToken->Tok.is(tok::numeric_constant)) {
995  CurrentToken->SpacesRequiredBefore = 1;
996  return Type;
997  }
998  // Hashes in the middle of a line can lead to any strange token
999  // sequence.
1000  if (!CurrentToken->Tok.getIdentifierInfo())
1001  return Type;
1002  switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
1003  case tok::pp_include:
1004  case tok::pp_include_next:
1005  case tok::pp_import:
1006  next();
1007  parseIncludeDirective();
1008  Type = LT_ImportStatement;
1009  break;
1010  case tok::pp_error:
1011  case tok::pp_warning:
1012  parseWarningOrError();
1013  break;
1014  case tok::pp_pragma:
1015  parsePragma();
1016  break;
1017  case tok::pp_if:
1018  case tok::pp_elif:
1019  Contexts.back().IsExpression = true;
1020  parseLine();
1021  break;
1022  default:
1023  break;
1024  }
1025  while (CurrentToken) {
1026  FormatToken *Tok = CurrentToken;
1027  next();
1028  if (Tok->is(tok::l_paren))
1029  parseParens();
1030  else if (Tok->isOneOf(Keywords.kw___has_include,
1031  Keywords.kw___has_include_next))
1032  parseHasInclude();
1033  }
1034  return Type;
1035  }
1036 
1037 public:
1038  LineType parseLine() {
1039  NonTemplateLess.clear();
1040  if (CurrentToken->is(tok::hash))
1041  return parsePreprocessorDirective();
1042 
1043  // Directly allow to 'import <string-literal>' to support protocol buffer
1044  // definitions (github.com/google/protobuf) or missing "#" (either way we
1045  // should not break the line).
1046  IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
1047  if ((Style.Language == FormatStyle::LK_Java &&
1048  CurrentToken->is(Keywords.kw_package)) ||
1049  (Info && Info->getPPKeywordID() == tok::pp_import &&
1050  CurrentToken->Next &&
1051  CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
1052  tok::kw_static))) {
1053  next();
1054  parseIncludeDirective();
1055  return LT_ImportStatement;
1056  }
1057 
1058  // If this line starts and ends in '<' and '>', respectively, it is likely
1059  // part of "#define <a/b.h>".
1060  if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
1061  parseIncludeDirective();
1062  return LT_ImportStatement;
1063  }
1064 
1065  // In .proto files, top-level options are very similar to import statements
1066  // and should not be line-wrapped.
1067  if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
1068  CurrentToken->is(Keywords.kw_option)) {
1069  next();
1070  if (CurrentToken && CurrentToken->is(tok::identifier))
1071  return LT_ImportStatement;
1072  }
1073 
1074  bool KeywordVirtualFound = false;
1075  bool ImportStatement = false;
1076 
1077  // import {...} from '...';
1078  if (Style.Language == FormatStyle::LK_JavaScript &&
1079  CurrentToken->is(Keywords.kw_import))
1080  ImportStatement = true;
1081 
1082  while (CurrentToken) {
1083  if (CurrentToken->is(tok::kw_virtual))
1084  KeywordVirtualFound = true;
1085  if (Style.Language == FormatStyle::LK_JavaScript) {
1086  // export {...} from '...';
1087  // An export followed by "from 'some string';" is a re-export from
1088  // another module identified by a URI and is treated as a
1089  // LT_ImportStatement (i.e. prevent wraps on it for long URIs).
1090  // Just "export {...};" or "export class ..." should not be treated as
1091  // an import in this sense.
1092  if (Line.First->is(tok::kw_export) &&
1093  CurrentToken->is(Keywords.kw_from) && CurrentToken->Next &&
1094  CurrentToken->Next->isStringLiteral())
1095  ImportStatement = true;
1096  if (isClosureImportStatement(*CurrentToken))
1097  ImportStatement = true;
1098  }
1099  if (!consumeToken())
1100  return LT_Invalid;
1101  }
1102  if (KeywordVirtualFound)
1103  return LT_VirtualFunctionDecl;
1104  if (ImportStatement)
1105  return LT_ImportStatement;
1106 
1107  if (Line.startsWith(TT_ObjCMethodSpecifier)) {
1108  if (Contexts.back().FirstObjCSelectorName)
1109  Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
1110  Contexts.back().LongestObjCSelectorName;
1111  return LT_ObjCMethodDecl;
1112  }
1113 
1114  return LT_Other;
1115  }
1116 
1117 private:
1118  bool isClosureImportStatement(const FormatToken &Tok) {
1119  // FIXME: Closure-library specific stuff should not be hard-coded but be
1120  // configurable.
1121  return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
1122  Tok.Next->Next &&
1123  (Tok.Next->Next->TokenText == "module" ||
1124  Tok.Next->Next->TokenText == "provide" ||
1125  Tok.Next->Next->TokenText == "require" ||
1126  Tok.Next->Next->TokenText == "forwardDeclare") &&
1127  Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
1128  }
1129 
1130  void resetTokenMetadata(FormatToken *Token) {
1131  if (!Token)
1132  return;
1133 
1134  // Reset token type in case we have already looked at it and then
1135  // recovered from an error (e.g. failure to find the matching >).
1136  if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
1137  TT_FunctionLBrace, TT_ImplicitStringLiteral,
1138  TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
1139  TT_OverloadedOperator, TT_RegexLiteral,
1140  TT_TemplateString, TT_ObjCStringLiteral))
1141  CurrentToken->Type = TT_Unknown;
1142  CurrentToken->Role.reset();
1143  CurrentToken->MatchingParen = nullptr;
1144  CurrentToken->FakeLParens.clear();
1145  CurrentToken->FakeRParens = 0;
1146  }
1147 
1148  void next() {
1149  if (CurrentToken) {
1150  CurrentToken->NestingLevel = Contexts.size() - 1;
1151  CurrentToken->BindingStrength = Contexts.back().BindingStrength;
1152  modifyContext(*CurrentToken);
1153  determineTokenType(*CurrentToken);
1154  CurrentToken = CurrentToken->Next;
1155  }
1156 
1157  resetTokenMetadata(CurrentToken);
1158  }
1159 
1160  /// A struct to hold information valid in a specific context, e.g.
1161  /// a pair of parenthesis.
1162  struct Context {
1163  Context(tok::TokenKind ContextKind, unsigned BindingStrength,
1164  bool IsExpression)
1165  : ContextKind(ContextKind), BindingStrength(BindingStrength),
1166  IsExpression(IsExpression) {}
1167 
1172  bool ColonIsForRangeExpr = false;
1173  bool ColonIsDictLiteral = false;
1175  FormatToken *FirstObjCSelectorName = nullptr;
1176  FormatToken *FirstStartOfName = nullptr;
1177  bool CanBeExpression = true;
1178  bool InTemplateArgument = false;
1179  bool InCtorInitializer = false;
1180  bool InInheritanceList = false;
1181  bool CaretFound = false;
1182  bool IsForEachMacro = false;
1184  };
1185 
1186  /// Puts a new \c Context onto the stack \c Contexts for the lifetime
1187  /// of each instance.
1188  struct ScopedContextCreator {
1189  AnnotatingParser &P;
1190 
1191  ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
1192  unsigned Increase)
1193  : P(P) {
1194  P.Contexts.push_back(Context(ContextKind,
1195  P.Contexts.back().BindingStrength + Increase,
1196  P.Contexts.back().IsExpression));
1197  }
1198 
1199  ~ScopedContextCreator() { P.Contexts.pop_back(); }
1200  };
1201 
1202  void modifyContext(const FormatToken &Current) {
1203  if (Current.getPrecedence() == prec::Assignment &&
1204  !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
1205  // Type aliases use `type X = ...;` in TypeScript and can be exported
1206  // using `export type ...`.
1207  !(Style.Language == FormatStyle::LK_JavaScript &&
1208  (Line.startsWith(Keywords.kw_type, tok::identifier) ||
1209  Line.startsWith(tok::kw_export, Keywords.kw_type,
1210  tok::identifier))) &&
1211  (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
1212  Contexts.back().IsExpression = true;
1213  if (!Line.startsWith(TT_UnaryOperator)) {
1214  for (FormatToken *Previous = Current.Previous;
1215  Previous && Previous->Previous &&
1216  !Previous->Previous->isOneOf(tok::comma, tok::semi);
1217  Previous = Previous->Previous) {
1218  if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
1219  Previous = Previous->MatchingParen;
1220  if (!Previous)
1221  break;
1222  }
1223  if (Previous->opensScope())
1224  break;
1225  if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
1226  Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
1227  Previous->Previous && Previous->Previous->isNot(tok::equal))
1228  Previous->Type = TT_PointerOrReference;
1229  }
1230  }
1231  } else if (Current.is(tok::lessless) &&
1232  (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
1233  Contexts.back().IsExpression = true;
1234  } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
1235  Contexts.back().IsExpression = true;
1236  } else if (Current.is(TT_TrailingReturnArrow)) {
1237  Contexts.back().IsExpression = false;
1238  } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
1239  Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
1240  } else if (Current.Previous &&
1241  Current.Previous->is(TT_CtorInitializerColon)) {
1242  Contexts.back().IsExpression = true;
1243  Contexts.back().InCtorInitializer = true;
1244  } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) {
1245  Contexts.back().InInheritanceList = true;
1246  } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
1247  for (FormatToken *Previous = Current.Previous;
1248  Previous && Previous->isOneOf(tok::star, tok::amp);
1249  Previous = Previous->Previous)
1250  Previous->Type = TT_PointerOrReference;
1251  if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
1252  Contexts.back().IsExpression = false;
1253  } else if (Current.is(tok::kw_new)) {
1254  Contexts.back().CanBeExpression = false;
1255  } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
1256  // This should be the condition or increment in a for-loop.
1257  Contexts.back().IsExpression = true;
1258  }
1259  }
1260 
1261  void determineTokenType(FormatToken &Current) {
1262  if (!Current.is(TT_Unknown))
1263  // The token type is already known.
1264  return;
1265 
1266  if (Style.Language == FormatStyle::LK_JavaScript) {
1267  if (Current.is(tok::exclaim)) {
1268  if (Current.Previous &&
1269  (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace,
1270  tok::r_paren, tok::r_square,
1271  tok::r_brace) ||
1272  Current.Previous->Tok.isLiteral())) {
1273  Current.Type = TT_JsNonNullAssertion;
1274  return;
1275  }
1276  if (Current.Next &&
1277  Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) {
1278  Current.Type = TT_JsNonNullAssertion;
1279  return;
1280  }
1281  }
1282  }
1283 
1284  // Line.MightBeFunctionDecl can only be true after the parentheses of a
1285  // function declaration have been found. In this case, 'Current' is a
1286  // trailing token of this declaration and thus cannot be a name.
1287  if (Current.is(Keywords.kw_instanceof)) {
1288  Current.Type = TT_BinaryOperator;
1289  } else if (isStartOfName(Current) &&
1290  (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
1291  Contexts.back().FirstStartOfName = &Current;
1292  Current.Type = TT_StartOfName;
1293  } else if (Current.is(tok::semi)) {
1294  // Reset FirstStartOfName after finding a semicolon so that a for loop
1295  // with multiple increment statements is not confused with a for loop
1296  // having multiple variable declarations.
1297  Contexts.back().FirstStartOfName = nullptr;
1298  } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
1299  AutoFound = true;
1300  } else if (Current.is(tok::arrow) &&
1301  Style.Language == FormatStyle::LK_Java) {
1302  Current.Type = TT_LambdaArrow;
1303  } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
1304  Current.NestingLevel == 0) {
1305  Current.Type = TT_TrailingReturnArrow;
1306  } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
1307  Current.Type = determineStarAmpUsage(Current,
1308  Contexts.back().CanBeExpression &&
1309  Contexts.back().IsExpression,
1310  Contexts.back().InTemplateArgument);
1311  } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
1312  Current.Type = determinePlusMinusCaretUsage(Current);
1313  if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
1314  Contexts.back().CaretFound = true;
1315  } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
1316  Current.Type = determineIncrementUsage(Current);
1317  } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
1318  Current.Type = TT_UnaryOperator;
1319  } else if (Current.is(tok::question)) {
1320  if (Style.Language == FormatStyle::LK_JavaScript &&
1321  Line.MustBeDeclaration && !Contexts.back().IsExpression) {
1322  // In JavaScript, `interface X { foo?(): bar; }` is an optional method
1323  // on the interface, not a ternary expression.
1324  Current.Type = TT_JsTypeOptionalQuestion;
1325  } else {
1326  Current.Type = TT_ConditionalExpr;
1327  }
1328  } else if (Current.isBinaryOperator() &&
1329  (!Current.Previous || Current.Previous->isNot(tok::l_square)) &&
1330  (!Current.is(tok::greater) &&
1331  Style.Language != FormatStyle::LK_TextProto)) {
1332  Current.Type = TT_BinaryOperator;
1333  } else if (Current.is(tok::comment)) {
1334  if (Current.TokenText.startswith("/*")) {
1335  if (Current.TokenText.endswith("*/"))
1336  Current.Type = TT_BlockComment;
1337  else
1338  // The lexer has for some reason determined a comment here. But we
1339  // cannot really handle it, if it isn't properly terminated.
1340  Current.Tok.setKind(tok::unknown);
1341  } else {
1342  Current.Type = TT_LineComment;
1343  }
1344  } else if (Current.is(tok::r_paren)) {
1345  if (rParenEndsCast(Current))
1346  Current.Type = TT_CastRParen;
1347  if (Current.MatchingParen && Current.Next &&
1348  !Current.Next->isBinaryOperator() &&
1349  !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace,
1350  tok::comma, tok::period, tok::arrow,
1351  tok::coloncolon))
1352  if (FormatToken *AfterParen = Current.MatchingParen->Next) {
1353  // Make sure this isn't the return type of an Obj-C block declaration
1354  if (AfterParen->Tok.isNot(tok::caret)) {
1355  if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
1356  if (BeforeParen->is(tok::identifier) &&
1357  BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
1358  (!BeforeParen->Previous ||
1359  BeforeParen->Previous->ClosesTemplateDeclaration))
1360  Current.Type = TT_FunctionAnnotationRParen;
1361  }
1362  }
1363  } else if (Current.is(tok::at) && Current.Next &&
1364  Style.Language != FormatStyle::LK_JavaScript &&
1365  Style.Language != FormatStyle::LK_Java) {
1366  // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it
1367  // marks declarations and properties that need special formatting.
1368  switch (Current.Next->Tok.getObjCKeywordID()) {
1369  case tok::objc_interface:
1370  case tok::objc_implementation:
1371  case tok::objc_protocol:
1372  Current.Type = TT_ObjCDecl;
1373  break;
1374  case tok::objc_property:
1375  Current.Type = TT_ObjCProperty;
1376  break;
1377  default:
1378  break;
1379  }
1380  } else if (Current.is(tok::period)) {
1381  FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1382  if (PreviousNoComment &&
1383  PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1384  Current.Type = TT_DesignatedInitializerPeriod;
1385  else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1386  Current.Previous->isOneOf(TT_JavaAnnotation,
1387  TT_LeadingJavaAnnotation)) {
1388  Current.Type = Current.Previous->Type;
1389  }
1390  } else if (canBeObjCSelectorComponent(Current) &&
1391  // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
1392  Current.Previous && Current.Previous->is(TT_CastRParen) &&
1393  Current.Previous->MatchingParen &&
1394  Current.Previous->MatchingParen->Previous &&
1395  Current.Previous->MatchingParen->Previous->is(
1396  TT_ObjCMethodSpecifier)) {
1397  // This is the first part of an Objective-C selector name. (If there's no
1398  // colon after this, this is the only place which annotates the identifier
1399  // as a selector.)
1400  Current.Type = TT_SelectorName;
1401  } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
1402  Current.Previous &&
1403  !Current.Previous->isOneOf(tok::equal, tok::at) &&
1404  Line.MightBeFunctionDecl && Contexts.size() == 1) {
1405  // Line.MightBeFunctionDecl can only be true after the parentheses of a
1406  // function declaration have been found.
1407  Current.Type = TT_TrailingAnnotation;
1408  } else if ((Style.Language == FormatStyle::LK_Java ||
1409  Style.Language == FormatStyle::LK_JavaScript) &&
1410  Current.Previous) {
1411  if (Current.Previous->is(tok::at) &&
1412  Current.isNot(Keywords.kw_interface)) {
1413  const FormatToken &AtToken = *Current.Previous;
1414  const FormatToken *Previous = AtToken.getPreviousNonComment();
1415  if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1416  Current.Type = TT_LeadingJavaAnnotation;
1417  else
1418  Current.Type = TT_JavaAnnotation;
1419  } else if (Current.Previous->is(tok::period) &&
1420  Current.Previous->isOneOf(TT_JavaAnnotation,
1421  TT_LeadingJavaAnnotation)) {
1422  Current.Type = Current.Previous->Type;
1423  }
1424  }
1425  }
1426 
1427  /// Take a guess at whether \p Tok starts a name of a function or
1428  /// variable declaration.
1429  ///
1430  /// This is a heuristic based on whether \p Tok is an identifier following
1431  /// something that is likely a type.
1432  bool isStartOfName(const FormatToken &Tok) {
1433  if (Tok.isNot(tok::identifier) || !Tok.Previous)
1434  return false;
1435 
1436  if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof,
1437  Keywords.kw_as))
1438  return false;
1439  if (Style.Language == FormatStyle::LK_JavaScript &&
1440  Tok.Previous->is(Keywords.kw_in))
1441  return false;
1442 
1443  // Skip "const" as it does not have an influence on whether this is a name.
1444  FormatToken *PreviousNotConst = Tok.getPreviousNonComment();
1445  while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1446  PreviousNotConst = PreviousNotConst->getPreviousNonComment();
1447 
1448  if (!PreviousNotConst)
1449  return false;
1450 
1451  bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1452  PreviousNotConst->Previous &&
1453  PreviousNotConst->Previous->is(tok::hash);
1454 
1455  if (PreviousNotConst->is(TT_TemplateCloser))
1456  return PreviousNotConst && PreviousNotConst->MatchingParen &&
1457  PreviousNotConst->MatchingParen->Previous &&
1458  PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1459  PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1460 
1461  if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1462  PreviousNotConst->MatchingParen->Previous &&
1463  PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1464  return true;
1465 
1466  return (!IsPPKeyword &&
1467  PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1468  PreviousNotConst->is(TT_PointerOrReference) ||
1469  PreviousNotConst->isSimpleTypeSpecifier();
1470  }
1471 
1472  /// Determine whether ')' is ending a cast.
1473  bool rParenEndsCast(const FormatToken &Tok) {
1474  // C-style casts are only used in C++ and Java.
1475  if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java)
1476  return false;
1477 
1478  // Empty parens aren't casts and there are no casts at the end of the line.
1479  if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1480  return false;
1481 
1482  FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1483  if (LeftOfParens) {
1484  // If there is a closing parenthesis left of the current parentheses,
1485  // look past it as these might be chained casts.
1486  if (LeftOfParens->is(tok::r_paren)) {
1487  if (!LeftOfParens->MatchingParen ||
1488  !LeftOfParens->MatchingParen->Previous)
1489  return false;
1490  LeftOfParens = LeftOfParens->MatchingParen->Previous;
1491  }
1492 
1493  // If there is an identifier (or with a few exceptions a keyword) right
1494  // before the parentheses, this is unlikely to be a cast.
1495  if (LeftOfParens->Tok.getIdentifierInfo() &&
1496  !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1497  tok::kw_delete))
1498  return false;
1499 
1500  // Certain other tokens right before the parentheses are also signals that
1501  // this cannot be a cast.
1502  if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1503  TT_TemplateCloser, tok::ellipsis))
1504  return false;
1505  }
1506 
1507  if (Tok.Next->is(tok::question))
1508  return false;
1509 
1510  // As Java has no function types, a "(" after the ")" likely means that this
1511  // is a cast.
1512  if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1513  return true;
1514 
1515  // If a (non-string) literal follows, this is likely a cast.
1516  if (Tok.Next->isNot(tok::string_literal) &&
1517  (Tok.Next->Tok.isLiteral() ||
1518  Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1519  return true;
1520 
1521  // Heuristically try to determine whether the parentheses contain a type.
1522  bool ParensAreType =
1523  !Tok.Previous ||
1524  Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1525  Tok.Previous->isSimpleTypeSpecifier();
1526  bool ParensCouldEndDecl =
1527  Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1528  if (ParensAreType && !ParensCouldEndDecl)
1529  return true;
1530 
1531  // At this point, we heuristically assume that there are no casts at the
1532  // start of the line. We assume that we have found most cases where there
1533  // are by the logic above, e.g. "(void)x;".
1534  if (!LeftOfParens)
1535  return false;
1536 
1537  // Certain token types inside the parentheses mean that this can't be a
1538  // cast.
1539  for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok;
1540  Token = Token->Next)
1541  if (Token->is(TT_BinaryOperator))
1542  return false;
1543 
1544  // If the following token is an identifier or 'this', this is a cast. All
1545  // cases where this can be something else are handled above.
1546  if (Tok.Next->isOneOf(tok::identifier, tok::kw_this))
1547  return true;
1548 
1549  if (!Tok.Next->Next)
1550  return false;
1551 
1552  // If the next token after the parenthesis is a unary operator, assume
1553  // that this is cast, unless there are unexpected tokens inside the
1554  // parenthesis.
1555  bool NextIsUnary =
1556  Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1557  if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1558  !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1559  return false;
1560  // Search for unexpected tokens.
1561  for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1562  Prev = Prev->Previous) {
1563  if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1564  return false;
1565  }
1566  return true;
1567  }
1568 
1569  /// Return the type of the given token assuming it is * or &.
1570  TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1571  bool InTemplateArgument) {
1572  if (Style.Language == FormatStyle::LK_JavaScript)
1573  return TT_BinaryOperator;
1574 
1575  const FormatToken *PrevToken = Tok.getPreviousNonComment();
1576  if (!PrevToken)
1577  return TT_UnaryOperator;
1578 
1579  const FormatToken *NextToken = Tok.getNextNonComment();
1580  if (!NextToken ||
1581  NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) ||
1582  (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1583  return TT_PointerOrReference;
1584 
1585  if (PrevToken->is(tok::coloncolon))
1586  return TT_PointerOrReference;
1587 
1588  if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1589  tok::comma, tok::semi, tok::kw_return, tok::colon,
1590  tok::equal, tok::kw_delete, tok::kw_sizeof,
1591  tok::kw_throw) ||
1592  PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1593  TT_UnaryOperator, TT_CastRParen))
1594  return TT_UnaryOperator;
1595 
1596  if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1597  return TT_PointerOrReference;
1598  if (NextToken->is(tok::kw_operator) && !IsExpression)
1599  return TT_PointerOrReference;
1600  if (NextToken->isOneOf(tok::comma, tok::semi))
1601  return TT_PointerOrReference;
1602 
1603  if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen) {
1604  FormatToken *TokenBeforeMatchingParen =
1605  PrevToken->MatchingParen->getPreviousNonComment();
1606  if (TokenBeforeMatchingParen &&
1607  TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype))
1608  return TT_PointerOrReference;
1609  }
1610 
1611  if (PrevToken->Tok.isLiteral() ||
1612  PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1613  tok::kw_false, tok::r_brace) ||
1614  NextToken->Tok.isLiteral() ||
1615  NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1616  NextToken->isUnaryOperator() ||
1617  // If we know we're in a template argument, there are no named
1618  // declarations. Thus, having an identifier on the right-hand side
1619  // indicates a binary operator.
1620  (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1621  return TT_BinaryOperator;
1622 
1623  // "&&(" is quite unlikely to be two successive unary "&".
1624  if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1625  return TT_BinaryOperator;
1626 
1627  // This catches some cases where evaluation order is used as control flow:
1628  // aaa && aaa->f();
1629  const FormatToken *NextNextToken = NextToken->getNextNonComment();
1630  if (NextNextToken && NextNextToken->is(tok::arrow))
1631  return TT_BinaryOperator;
1632 
1633  // It is very unlikely that we are going to find a pointer or reference type
1634  // definition on the RHS of an assignment.
1635  if (IsExpression && !Contexts.back().CaretFound)
1636  return TT_BinaryOperator;
1637 
1638  return TT_PointerOrReference;
1639  }
1640 
1641  TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1642  const FormatToken *PrevToken = Tok.getPreviousNonComment();
1643  if (!PrevToken)
1644  return TT_UnaryOperator;
1645 
1646  if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator))
1647  // This must be a sequence of leading unary operators.
1648  return TT_UnaryOperator;
1649 
1650  // Use heuristics to recognize unary operators.
1651  if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1652  tok::question, tok::colon, tok::kw_return,
1653  tok::kw_case, tok::at, tok::l_brace))
1654  return TT_UnaryOperator;
1655 
1656  // There can't be two consecutive binary operators.
1657  if (PrevToken->is(TT_BinaryOperator))
1658  return TT_UnaryOperator;
1659 
1660  // Fall back to marking the token as binary operator.
1661  return TT_BinaryOperator;
1662  }
1663 
1664  /// Determine whether ++/-- are pre- or post-increments/-decrements.
1665  TokenType determineIncrementUsage(const FormatToken &Tok) {
1666  const FormatToken *PrevToken = Tok.getPreviousNonComment();
1667  if (!PrevToken || PrevToken->is(TT_CastRParen))
1668  return TT_UnaryOperator;
1669  if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1670  return TT_TrailingUnaryOperator;
1671 
1672  return TT_UnaryOperator;
1673  }
1674 
1675  SmallVector<Context, 8> Contexts;
1676 
1677  const FormatStyle &Style;
1678  AnnotatedLine &Line;
1679  FormatToken *CurrentToken;
1680  bool AutoFound;
1681  const AdditionalKeywords &Keywords;
1682 
1683  // Set of "<" tokens that do not open a template parameter list. If parseAngle
1684  // determines that a specific token can't be a template opener, it will make
1685  // same decision irrespective of the decisions for tokens leading up to it.
1686  // Store this information to prevent this from causing exponential runtime.
1687  llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1688 };
1689 
1690 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1691 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1692 
1693 /// Parses binary expressions by inserting fake parenthesis based on
1694 /// operator precedence.
1695 class ExpressionParser {
1696 public:
1697  ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1698  AnnotatedLine &Line)
1699  : Style(Style), Keywords(Keywords), Current(Line.First) {}
1700 
1701  /// Parse expressions with the given operator precedence.
1702  void parse(int Precedence = 0) {
1703  // Skip 'return' and ObjC selector colons as they are not part of a binary
1704  // expression.
1705  while (Current && (Current->is(tok::kw_return) ||
1706  (Current->is(tok::colon) &&
1707  Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1708  next();
1709 
1710  if (!Current || Precedence > PrecedenceArrowAndPeriod)
1711  return;
1712 
1713  // Conditional expressions need to be parsed separately for proper nesting.
1714  if (Precedence == prec::Conditional) {
1715  parseConditionalExpr();
1716  return;
1717  }
1718 
1719  // Parse unary operators, which all have a higher precedence than binary
1720  // operators.
1721  if (Precedence == PrecedenceUnaryOperator) {
1722  parseUnaryOperator();
1723  return;
1724  }
1725 
1726  FormatToken *Start = Current;
1727  FormatToken *LatestOperator = nullptr;
1728  unsigned OperatorIndex = 0;
1729 
1730  while (Current) {
1731  // Consume operators with higher precedence.
1732  parse(Precedence + 1);
1733 
1734  int CurrentPrecedence = getCurrentPrecedence();
1735 
1736  if (Current && Current->is(TT_SelectorName) &&
1737  Precedence == CurrentPrecedence) {
1738  if (LatestOperator)
1739  addFakeParenthesis(Start, prec::Level(Precedence));
1740  Start = Current;
1741  }
1742 
1743  // At the end of the line or when an operator with higher precedence is
1744  // found, insert fake parenthesis and return.
1745  if (!Current ||
1746  (Current->closesScope() &&
1747  (Current->MatchingParen || Current->is(TT_TemplateString))) ||
1748  (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1749  (CurrentPrecedence == prec::Conditional &&
1750  Precedence == prec::Assignment && Current->is(tok::colon))) {
1751  break;
1752  }
1753 
1754  // Consume scopes: (), [], <> and {}
1755  if (Current->opensScope()) {
1756  // In fragment of a JavaScript template string can look like '}..${' and
1757  // thus close a scope and open a new one at the same time.
1758  while (Current && (!Current->closesScope() || Current->opensScope())) {
1759  next();
1760  parse();
1761  }
1762  next();
1763  } else {
1764  // Operator found.
1765  if (CurrentPrecedence == Precedence) {
1766  if (LatestOperator)
1767  LatestOperator->NextOperator = Current;
1768  LatestOperator = Current;
1769  Current->OperatorIndex = OperatorIndex;
1770  ++OperatorIndex;
1771  }
1772  next(/*SkipPastLeadingComments=*/Precedence > 0);
1773  }
1774  }
1775 
1776  if (LatestOperator && (Current || Precedence > 0)) {
1777  // LatestOperator->LastOperator = true;
1778  if (Precedence == PrecedenceArrowAndPeriod) {
1779  // Call expressions don't have a binary operator precedence.
1780  addFakeParenthesis(Start, prec::Unknown);
1781  } else {
1782  addFakeParenthesis(Start, prec::Level(Precedence));
1783  }
1784  }
1785  }
1786 
1787 private:
1788  /// Gets the precedence (+1) of the given token for binary operators
1789  /// and other tokens that we treat like binary operators.
1790  int getCurrentPrecedence() {
1791  if (Current) {
1792  const FormatToken *NextNonComment = Current->getNextNonComment();
1793  if (Current->is(TT_ConditionalExpr))
1794  return prec::Conditional;
1795  if (NextNonComment && Current->is(TT_SelectorName) &&
1796  (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) ||
1797  ((Style.Language == FormatStyle::LK_Proto ||
1798  Style.Language == FormatStyle::LK_TextProto) &&
1799  NextNonComment->is(tok::less))))
1800  return prec::Assignment;
1801  if (Current->is(TT_JsComputedPropertyName))
1802  return prec::Assignment;
1803  if (Current->is(TT_LambdaArrow))
1804  return prec::Comma;
1805  if (Current->is(TT_JsFatArrow))
1806  return prec::Assignment;
1807  if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) ||
1808  (Current->is(tok::comment) && NextNonComment &&
1809  NextNonComment->is(TT_SelectorName)))
1810  return 0;
1811  if (Current->is(TT_RangeBasedForLoopColon))
1812  return prec::Comma;
1813  if ((Style.Language == FormatStyle::LK_Java ||
1814  Style.Language == FormatStyle::LK_JavaScript) &&
1815  Current->is(Keywords.kw_instanceof))
1816  return prec::Relational;
1817  if (Style.Language == FormatStyle::LK_JavaScript &&
1818  Current->isOneOf(Keywords.kw_in, Keywords.kw_as))
1819  return prec::Relational;
1820  if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1821  return Current->getPrecedence();
1822  if (Current->isOneOf(tok::period, tok::arrow))
1823  return PrecedenceArrowAndPeriod;
1824  if ((Style.Language == FormatStyle::LK_Java ||
1825  Style.Language == FormatStyle::LK_JavaScript) &&
1826  Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1827  Keywords.kw_throws))
1828  return 0;
1829  }
1830  return -1;
1831  }
1832 
1833  void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1834  Start->FakeLParens.push_back(Precedence);
1835  if (Precedence > prec::Unknown)
1836  Start->StartsBinaryExpression = true;
1837  if (Current) {
1838  FormatToken *Previous = Current->Previous;
1839  while (Previous->is(tok::comment) && Previous->Previous)
1840  Previous = Previous->Previous;
1841  ++Previous->FakeRParens;
1842  if (Precedence > prec::Unknown)
1843  Previous->EndsBinaryExpression = true;
1844  }
1845  }
1846 
1847  /// Parse unary operator expressions and surround them with fake
1848  /// parentheses if appropriate.
1849  void parseUnaryOperator() {
1851  while (Current && Current->is(TT_UnaryOperator)) {
1852  Tokens.push_back(Current);
1853  next();
1854  }
1855  parse(PrecedenceArrowAndPeriod);
1856  for (FormatToken *Token : llvm::reverse(Tokens))
1857  // The actual precedence doesn't matter.
1858  addFakeParenthesis(Token, prec::Unknown);
1859  }
1860 
1861  void parseConditionalExpr() {
1862  while (Current && Current->isTrailingComment()) {
1863  next();
1864  }
1865  FormatToken *Start = Current;
1866  parse(prec::LogicalOr);
1867  if (!Current || !Current->is(tok::question))
1868  return;
1869  next();
1870  parse(prec::Assignment);
1871  if (!Current || Current->isNot(TT_ConditionalExpr))
1872  return;
1873  next();
1874  parse(prec::Assignment);
1875  addFakeParenthesis(Start, prec::Conditional);
1876  }
1877 
1878  void next(bool SkipPastLeadingComments = true) {
1879  if (Current)
1880  Current = Current->Next;
1881  while (Current &&
1882  (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1883  Current->isTrailingComment())
1884  Current = Current->Next;
1885  }
1886 
1887  const FormatStyle &Style;
1888  const AdditionalKeywords &Keywords;
1889  FormatToken *Current;
1890 };
1891 
1892 } // end anonymous namespace
1893 
1896  const AnnotatedLine *NextNonCommentLine = nullptr;
1898  E = Lines.rend();
1899  I != E; ++I) {
1900  bool CommentLine = true;
1901  for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) {
1902  if (!Tok->is(tok::comment)) {
1903  CommentLine = false;
1904  break;
1905  }
1906  }
1907 
1908  // If the comment is currently aligned with the line immediately following
1909  // it, that's probably intentional and we should keep it.
1910  if (NextNonCommentLine && CommentLine &&
1911  NextNonCommentLine->First->NewlinesBefore <= 1 &&
1912  NextNonCommentLine->First->OriginalColumn ==
1913  (*I)->First->OriginalColumn) {
1914  // Align comments for preprocessor lines with the # in column 0.
1915  // Otherwise, align with the next line.
1916  (*I)->Level = (NextNonCommentLine->Type == LT_PreprocessorDirective ||
1917  NextNonCommentLine->Type == LT_ImportStatement)
1918  ? 0
1919  : NextNonCommentLine->Level;
1920  } else {
1921  NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1922  }
1923 
1924  setCommentLineLevels((*I)->Children);
1925  }
1926 }
1927 
1928 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
1929  unsigned Result = 0;
1930  for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
1931  Result = std::max(Result, Tok->NestingLevel);
1932  return Result;
1933 }
1934 
1937  E = Line.Children.end();
1938  I != E; ++I) {
1939  annotate(**I);
1940  }
1941  AnnotatingParser Parser(Style, Line, Keywords);
1942  Line.Type = Parser.parseLine();
1943 
1944  // With very deep nesting, ExpressionParser uses lots of stack and the
1945  // formatting algorithm is very slow. We're not going to do a good job here
1946  // anyway - it's probably generated code being formatted by mistake.
1947  // Just skip the whole line.
1948  if (maxNestingDepth(Line) > 50)
1949  Line.Type = LT_Invalid;
1950 
1951  if (Line.Type == LT_Invalid)
1952  return;
1953 
1954  ExpressionParser ExprParser(Style, Keywords, Line);
1955  ExprParser.parse();
1956 
1957  if (Line.startsWith(TT_ObjCMethodSpecifier))
1958  Line.Type = LT_ObjCMethodDecl;
1959  else if (Line.startsWith(TT_ObjCDecl))
1960  Line.Type = LT_ObjCDecl;
1961  else if (Line.startsWith(TT_ObjCProperty))
1962  Line.Type = LT_ObjCProperty;
1963 
1964  Line.First->SpacesRequiredBefore = 1;
1965  Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1966 }
1967 
1968 // This function heuristically determines whether 'Current' starts the name of a
1969 // function declaration.
1970 static bool isFunctionDeclarationName(const FormatToken &Current,
1971  const AnnotatedLine &Line) {
1972  auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * {
1973  for (; Next; Next = Next->Next) {
1974  if (Next->is(TT_OverloadedOperatorLParen))
1975  return Next;
1976  if (Next->is(TT_OverloadedOperator))
1977  continue;
1978  if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1979  // For 'new[]' and 'delete[]'.
1980  if (Next->Next && Next->Next->is(tok::l_square) && Next->Next->Next &&
1981  Next->Next->Next->is(tok::r_square))
1982  Next = Next->Next->Next;
1983  continue;
1984  }
1985 
1986  break;
1987  }
1988  return nullptr;
1989  };
1990 
1991  // Find parentheses of parameter list.
1992  const FormatToken *Next = Current.Next;
1993  if (Current.is(tok::kw_operator)) {
1994  if (Current.Previous && Current.Previous->is(tok::coloncolon))
1995  return false;
1996  Next = skipOperatorName(Next);
1997  } else {
1998  if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1999  return false;
2000  for (; Next; Next = Next->Next) {
2001  if (Next->is(TT_TemplateOpener)) {
2002  Next = Next->MatchingParen;
2003  } else if (Next->is(tok::coloncolon)) {
2004  Next = Next->Next;
2005  if (!Next)
2006  return false;
2007  if (Next->is(tok::kw_operator)) {
2008  Next = skipOperatorName(Next->Next);
2009  break;
2010  }
2011  if (!Next->is(tok::identifier))
2012  return false;
2013  } else if (Next->is(tok::l_paren)) {
2014  break;
2015  } else {
2016  return false;
2017  }
2018  }
2019  }
2020 
2021  // Check whether parameter list can belong to a function declaration.
2022  if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
2023  return false;
2024  // If the lines ends with "{", this is likely an function definition.
2025  if (Line.Last->is(tok::l_brace))
2026  return true;
2027  if (Next->Next == Next->MatchingParen)
2028  return true; // Empty parentheses.
2029  // If there is an &/&& after the r_paren, this is likely a function.
2030  if (Next->MatchingParen->Next &&
2031  Next->MatchingParen->Next->is(TT_PointerOrReference))
2032  return true;
2033  for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
2034  Tok = Tok->Next) {
2035  if (Tok->is(tok::l_paren) && Tok->MatchingParen) {
2036  Tok = Tok->MatchingParen;
2037  continue;
2038  }
2039  if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
2040  Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
2041  return true;
2042  if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
2043  Tok->Tok.isLiteral())
2044  return false;
2045  }
2046  return false;
2047 }
2048 
2049 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
2050  assert(Line.MightBeFunctionDecl);
2051 
2055  Line.Level > 0)
2056  return false;
2057 
2060  return false;
2061  case FormatStyle::RTBS_All:
2063  return true;
2066  return Line.mightBeFunctionDefinition();
2067  }
2068 
2069  return false;
2070 }
2071 
2074  E = Line.Children.end();
2075  I != E; ++I) {
2076  calculateFormattingInformation(**I);
2077  }
2078 
2079  Line.First->TotalLength =
2081  : Line.FirstStartColumn + Line.First->ColumnWidth;
2082  FormatToken *Current = Line.First->Next;
2083  bool InFunctionDecl = Line.MightBeFunctionDecl;
2084  while (Current) {
2085  if (isFunctionDeclarationName(*Current, Line))
2086  Current->Type = TT_FunctionDeclarationName;
2087  if (Current->is(TT_LineComment)) {
2088  if (Current->Previous->BlockKind == BK_BracedInit &&
2089  Current->Previous->opensScope())
2090  Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
2091  else
2093 
2094  // If we find a trailing comment, iterate backwards to determine whether
2095  // it seems to relate to a specific parameter. If so, break before that
2096  // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
2097  // to the previous line in:
2098  // SomeFunction(a,
2099  // b, // comment
2100  // c);
2101  if (!Current->HasUnescapedNewline) {
2102  for (FormatToken *Parameter = Current->Previous; Parameter;
2103  Parameter = Parameter->Previous) {
2104  if (Parameter->isOneOf(tok::comment, tok::r_brace))
2105  break;
2106  if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
2107  if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
2108  Parameter->HasUnescapedNewline)
2109  Parameter->MustBreakBefore = true;
2110  break;
2111  }
2112  }
2113  }
2114  } else if (Current->SpacesRequiredBefore == 0 &&
2115  spaceRequiredBefore(Line, *Current)) {
2116  Current->SpacesRequiredBefore = 1;
2117  }
2118 
2119  Current->MustBreakBefore =
2120  Current->MustBreakBefore || mustBreakBefore(Line, *Current);
2121 
2122  if (!Current->MustBreakBefore && InFunctionDecl &&
2123  Current->is(TT_FunctionDeclarationName))
2124  Current->MustBreakBefore = mustBreakForReturnType(Line);
2125 
2126  Current->CanBreakBefore =
2127  Current->MustBreakBefore || canBreakBefore(Line, *Current);
2128  unsigned ChildSize = 0;
2129  if (Current->Previous->Children.size() == 1) {
2130  FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
2131  ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
2132  : LastOfChild.TotalLength + 1;
2133  }
2134  const FormatToken *Prev = Current->Previous;
2135  if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
2136  (Prev->Children.size() == 1 &&
2137  Prev->Children[0]->First->MustBreakBefore) ||
2138  Current->IsMultiline)
2139  Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
2140  else
2141  Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
2142  ChildSize + Current->SpacesRequiredBefore;
2143 
2144  if (Current->is(TT_CtorInitializerColon))
2145  InFunctionDecl = false;
2146 
2147  // FIXME: Only calculate this if CanBreakBefore is true once static
2148  // initializers etc. are sorted out.
2149  // FIXME: Move magic numbers to a better place.
2150 
2151  // Reduce penalty for aligning ObjC method arguments using the colon
2152  // alignment as this is the canonical way (still prefer fitting everything
2153  // into one line if possible). Trying to fit a whole expression into one
2154  // line should not force other line breaks (e.g. when ObjC method
2155  // expression is a part of other expression).
2156  Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl);
2158  Current->is(TT_SelectorName) && Current->ParameterIndex > 0) {
2159  if (Current->ParameterIndex == 1)
2160  Current->SplitPenalty += 5 * Current->BindingStrength;
2161  } else {
2162  Current->SplitPenalty += 20 * Current->BindingStrength;
2163  }
2164 
2165  Current = Current->Next;
2166  }
2167 
2168  calculateUnbreakableTailLengths(Line);
2169  unsigned IndentLevel = Line.Level;
2170  for (Current = Line.First; Current != nullptr; Current = Current->Next) {
2171  if (Current->Role)
2172  Current->Role->precomputeFormattingInfos(Current);
2173  if (Current->MatchingParen &&
2175  assert(IndentLevel > 0);
2176  --IndentLevel;
2177  }
2178  Current->IndentLevel = IndentLevel;
2179  if (Current->opensBlockOrBlockTypeList(Style))
2180  ++IndentLevel;
2181  }
2182 
2183  LLVM_DEBUG({ printDebugInfo(Line); });
2184 }
2185 
2186 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
2187  unsigned UnbreakableTailLength = 0;
2188  FormatToken *Current = Line.Last;
2189  while (Current) {
2190  Current->UnbreakableTailLength = UnbreakableTailLength;
2191  if (Current->CanBreakBefore ||
2192  Current->isOneOf(tok::comment, tok::string_literal)) {
2193  UnbreakableTailLength = 0;
2194  } else {
2195  UnbreakableTailLength +=
2196  Current->ColumnWidth + Current->SpacesRequiredBefore;
2197  }
2198  Current = Current->Previous;
2199  }
2200 }
2201 
2202 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
2203  const FormatToken &Tok,
2204  bool InFunctionDecl) {
2205  const FormatToken &Left = *Tok.Previous;
2206  const FormatToken &Right = Tok;
2207 
2208  if (Left.is(tok::semi))
2209  return 0;
2210 
2212  if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
2213  return 1;
2214  if (Right.is(Keywords.kw_implements))
2215  return 2;
2216  if (Left.is(tok::comma) && Left.NestingLevel == 0)
2217  return 3;
2218  } else if (Style.Language == FormatStyle::LK_JavaScript) {
2219  if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
2220  return 100;
2221  if (Left.is(TT_JsTypeColon))
2222  return 35;
2223  if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2224  (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2225  return 100;
2226  // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()".
2227  if (Left.opensScope() && Right.closesScope())
2228  return 200;
2229  }
2230 
2231  if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2232  return 1;
2233  if (Right.is(tok::l_square)) {
2235  return 1;
2236  if (Left.is(tok::r_square))
2237  return 200;
2238  // Slightly prefer formatting local lambda definitions like functions.
2239  if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
2240  return 35;
2241  if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
2242  TT_ArrayInitializerLSquare,
2243  TT_DesignatedInitializerLSquare, TT_AttributeSquare))
2244  return 500;
2245  }
2246 
2247  if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2248  Right.is(tok::kw_operator)) {
2249  if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
2250  return 3;
2251  if (Left.is(TT_StartOfName))
2252  return 110;
2253  if (InFunctionDecl && Right.NestingLevel == 0)
2255  return 200;
2256  }
2257  if (Right.is(TT_PointerOrReference))
2258  return 190;
2259  if (Right.is(TT_LambdaArrow))
2260  return 110;
2261  if (Left.is(tok::equal) && Right.is(tok::l_brace))
2262  return 160;
2263  if (Left.is(TT_CastRParen))
2264  return 100;
2265  if (Left.is(tok::coloncolon) ||
2266  (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
2267  return 500;
2268  if (Left.isOneOf(tok::kw_class, tok::kw_struct))
2269  return 5000;
2270  if (Left.is(tok::comment))
2271  return 1000;
2272 
2273  if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon,
2274  TT_CtorInitializerColon))
2275  return 2;
2276 
2277  if (Right.isMemberAccess()) {
2278  // Breaking before the "./->" of a chained call/member access is reasonably
2279  // cheap, as formatting those with one call per line is generally
2280  // desirable. In particular, it should be cheaper to break before the call
2281  // than it is to break inside a call's parameters, which could lead to weird
2282  // "hanging" indents. The exception is the very last "./->" to support this
2283  // frequent pattern:
2284  //
2285  // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
2286  // dddddddd);
2287  //
2288  // which might otherwise be blown up onto many lines. Here, clang-format
2289  // won't produce "hanging" indents anyway as there is no other trailing
2290  // call.
2291  //
2292  // Also apply higher penalty is not a call as that might lead to a wrapping
2293  // like:
2294  //
2295  // aaaaaaa
2296  // .aaaaaaaaa.bbbbbbbb(cccccccc);
2297  return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
2298  ? 150
2299  : 35;
2300  }
2301 
2302  if (Right.is(TT_TrailingAnnotation) &&
2303  (!Right.Next || Right.Next->isNot(tok::l_paren))) {
2304  // Moving trailing annotations to the next line is fine for ObjC method
2305  // declarations.
2306  if (Line.startsWith(TT_ObjCMethodSpecifier))
2307  return 10;
2308  // Generally, breaking before a trailing annotation is bad unless it is
2309  // function-like. It seems to be especially preferable to keep standard
2310  // annotations (i.e. "const", "final" and "override") on the same line.
2311  // Use a slightly higher penalty after ")" so that annotations like
2312  // "const override" are kept together.
2313  bool is_short_annotation = Right.TokenText.size() < 10;
2314  return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
2315  }
2316 
2317  // In for-loops, prefer breaking at ',' and ';'.
2318  if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
2319  return 4;
2320 
2321  // In Objective-C method expressions, prefer breaking before "param:" over
2322  // breaking after it.
2323  if (Right.is(TT_SelectorName))
2324  return 0;
2325  if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
2326  return Line.MightBeFunctionDecl ? 50 : 500;
2327 
2328  // In Objective-C type declarations, avoid breaking after the category's
2329  // open paren (we'll prefer breaking after the protocol list's opening
2330  // angle bracket, if present).
2331  if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous &&
2332  Left.Previous->isOneOf(tok::identifier, tok::greater))
2333  return 500;
2334 
2335  if (Left.is(tok::l_paren) && InFunctionDecl &&
2337  return 100;
2338  if (Left.is(tok::l_paren) && Left.Previous &&
2339  (Left.Previous->isOneOf(tok::kw_if, tok::kw_for) ||
2340  Left.Previous->endsSequence(tok::kw_constexpr, tok::kw_if)))
2341  return 1000;
2342  if (Left.is(tok::equal) && InFunctionDecl)
2343  return 110;
2344  if (Right.is(tok::r_brace))
2345  return 1;
2346  if (Left.is(TT_TemplateOpener))
2347  return 100;
2348  if (Left.opensScope()) {
2350  return 0;
2351  if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle)
2352  return 19;
2354  : 19;
2355  }
2356  if (Left.is(TT_JavaAnnotation))
2357  return 50;
2358 
2359  if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous &&
2360  Left.Previous->isLabelString() &&
2361  (Left.NextOperator || Left.OperatorIndex != 0))
2362  return 50;
2363  if (Right.is(tok::plus) && Left.isLabelString() &&
2364  (Right.NextOperator || Right.OperatorIndex != 0))
2365  return 25;
2366  if (Left.is(tok::comma))
2367  return 1;
2368  if (Right.is(tok::lessless) && Left.isLabelString() &&
2369  (Right.NextOperator || Right.OperatorIndex != 1))
2370  return 25;
2371  if (Right.is(tok::lessless)) {
2372  // Breaking at a << is really cheap.
2373  if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0)
2374  // Slightly prefer to break before the first one in log-like statements.
2375  return 2;
2376  return 1;
2377  }
2378  if (Left.ClosesTemplateDeclaration)
2380  if (Left.is(TT_ConditionalExpr))
2381  return prec::Conditional;
2382  prec::Level Level = Left.getPrecedence();
2383  if (Level == prec::Unknown)
2384  Level = Right.getPrecedence();
2385  if (Level == prec::Assignment)
2387  if (Level != prec::Unknown)
2388  return Level;
2389 
2390  return 3;
2391 }
2392 
2393 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2394  const FormatToken &Left,
2395  const FormatToken &Right) {
2396  if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2397  return true;
2398  if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java)
2399  return true;
2401  Left.Tok.getObjCKeywordID() == tok::objc_property)
2402  return true;
2403  if (Right.is(tok::hashhash))
2404  return Left.is(tok::hash);
2405  if (Left.isOneOf(tok::hashhash, tok::hash))
2406  return Right.is(tok::hash);
2407  if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
2409  if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2410  return (Right.is(TT_CastRParen) ||
2411  (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2414  if (Right.isOneOf(tok::semi, tok::comma))
2415  return false;
2416  if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) {
2417  bool IsLightweightGeneric =
2418  Right.MatchingParen && Right.MatchingParen->Next &&
2419  Right.MatchingParen->Next->is(tok::colon);
2420  return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList;
2421  }
2422  if (Right.is(tok::less) && Left.is(tok::kw_template))
2424  if (Left.isOneOf(tok::exclaim, tok::tilde))
2425  return false;
2426  if (Left.is(tok::at) &&
2427  Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2428  tok::numeric_constant, tok::l_paren, tok::l_brace,
2429  tok::kw_true, tok::kw_false))
2430  return false;
2431  if (Left.is(tok::colon))
2432  return !Left.is(TT_ObjCMethodExpr);
2433  if (Left.is(tok::coloncolon))
2434  return false;
2435  if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) {
2438  (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) {
2439  // Format empty list as `<>`.
2440  if (Left.is(tok::less) && Right.is(tok::greater))
2441  return false;
2442  return !Style.Cpp11BracedListStyle;
2443  }
2444  return false;
2445  }
2446  if (Right.is(tok::ellipsis))
2447  return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2448  Left.Previous->is(tok::kw_case));
2449  if (Left.is(tok::l_square) && Right.is(tok::amp))
2450  return false;
2451  if (Right.is(TT_PointerOrReference)) {
2452  if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) {
2453  if (!Left.MatchingParen)
2454  return true;
2455  FormatToken *TokenBeforeMatchingParen =
2457  if (!TokenBeforeMatchingParen ||
2458  !TokenBeforeMatchingParen->isOneOf(tok::kw_typeof, tok::kw_decltype))
2459  return true;
2460  }
2461  return (Left.Tok.isLiteral() ||
2462  (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
2464  (Line.IsMultiVariableDeclStmt &&
2465  (Left.NestingLevel == 0 ||
2466  (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
2467  }
2468  if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
2469  (!Left.is(TT_PointerOrReference) ||
2471  !Line.IsMultiVariableDeclStmt)))
2472  return true;
2473  if (Left.is(TT_PointerOrReference))
2474  return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
2475  (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) &&
2476  !Right.is(TT_StartOfName)) ||
2477  (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
2478  (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
2479  tok::l_paren) &&
2481  !Line.IsMultiVariableDeclStmt) &&
2482  Left.Previous &&
2483  !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
2484  if (Right.is(tok::star) && Left.is(tok::l_paren))
2485  return false;
2486  const auto SpaceRequiredForArrayInitializerLSquare =
2487  [](const FormatToken &LSquareTok, const FormatStyle &Style) {
2492  LSquareTok.endsSequence(tok::l_square, tok::colon,
2493  TT_SelectorName));
2494  };
2495  if (Left.is(tok::l_square))
2496  return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) &&
2497  SpaceRequiredForArrayInitializerLSquare(Left, Style)) ||
2498  (Left.isOneOf(TT_ArraySubscriptLSquare,
2499  TT_StructuredBindingLSquare) &&
2500  Style.SpacesInSquareBrackets && Right.isNot(tok::r_square));
2501  if (Right.is(tok::r_square))
2502  return Right.MatchingParen &&
2503  ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) &&
2504  SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen,
2505  Style)) ||
2507  Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare,
2508  TT_StructuredBindingLSquare)) ||
2509  Right.MatchingParen->is(TT_AttributeParen));
2510  if (Right.is(tok::l_square) &&
2511  !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
2512  TT_DesignatedInitializerLSquare,
2513  TT_StructuredBindingLSquare, TT_AttributeSquare) &&
2514  !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
2515  return false;
2516  if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
2517  return !Left.Children.empty(); // No spaces in "{}".
2518  if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
2519  (Right.is(tok::r_brace) && Right.MatchingParen &&
2520  Right.MatchingParen->BlockKind != BK_Block))
2521  return !Style.Cpp11BracedListStyle;
2522  if (Left.is(TT_BlockComment))
2523  // No whitespace in x(/*foo=*/1), except for JavaScript.
2525  !Left.TokenText.endswith("=*/");
2526  if (Right.is(tok::l_paren)) {
2527  if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) ||
2528  (Left.is(tok::r_square) && Left.is(TT_AttributeSquare)))
2529  return true;
2530  return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
2532  (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
2533  tok::kw_switch, tok::kw_case, TT_ForEachMacro,
2534  TT_ObjCForIn) ||
2535  Left.endsSequence(tok::kw_constexpr, tok::kw_if) ||
2536  (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
2537  tok::kw_new, tok::kw_delete) &&
2538  (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
2540  (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
2541  Left.is(tok::r_paren)) &&
2542  Line.Type != LT_PreprocessorDirective);
2543  }
2544  if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
2545  return false;
2546  if (Right.is(TT_UnaryOperator))
2547  return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2548  (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2549  if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2550  tok::r_paren) ||
2551  Left.isSimpleTypeSpecifier()) &&
2552  Right.is(tok::l_brace) && Right.getNextNonComment() &&
2553  Right.BlockKind != BK_Block)
2554  return false;
2555  if (Left.is(tok::period) || Right.is(tok::period))
2556  return false;
2557  if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2558  return false;
2559  if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2560  Left.MatchingParen->Previous &&
2561  (Left.MatchingParen->Previous->is(tok::period) ||
2562  Left.MatchingParen->Previous->is(tok::coloncolon)))
2563  // Java call to generic function with explicit type:
2564  // A.<B<C<...>>>DoSomething();
2565  // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference.
2566  return false;
2567  if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2568  return false;
2569  if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at))
2570  // Objective-C dictionary literal -> no space after opening brace.
2571  return false;
2572  if (Right.is(tok::r_brace) && Right.MatchingParen &&
2573  Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at))
2574  // Objective-C dictionary literal -> no space before closing brace.
2575  return false;
2576  return true;
2577 }
2578 
2579 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2580  const FormatToken &Right) {
2581  const FormatToken &Left = *Right.Previous;
2582  if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2583  return true; // Never ever merge two identifiers.
2584  if (Style.isCpp()) {
2585  if (Left.is(tok::kw_operator))
2586  return Right.is(tok::coloncolon);
2587  if (Right.is(tok::l_brace) && Right.BlockKind == BK_BracedInit &&
2589  return true;
2590  } else if (Style.Language == FormatStyle::LK_Proto ||
2592  if (Right.is(tok::period) &&
2593  Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2594  Keywords.kw_repeated, Keywords.kw_extend))
2595  return true;
2596  if (Right.is(tok::l_paren) &&
2597  Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2598  return true;
2599  if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName))
2600  return true;
2601  // Slashes occur in text protocol extension syntax: [type/type] { ... }.
2602  if (Left.is(tok::slash) || Right.is(tok::slash))
2603  return false;
2604  if (Left.MatchingParen && Left.MatchingParen->is(TT_ProtoExtensionLSquare) &&
2605  Right.isOneOf(tok::l_brace, tok::less))
2606  return !Style.Cpp11BracedListStyle;
2607  // A percent is probably part of a formatting specification, such as %lld.
2608  if (Left.is(tok::percent))
2609  return false;
2610  // Preserve the existence of a space before a percent for cases like 0x%04x
2611  // and "%d %d"
2612  if (Left.is(tok::numeric_constant) && Right.is(tok::percent))
2613  return Right.WhitespaceRange.getEnd() != Right.WhitespaceRange.getBegin();
2614  } else if (Style.Language == FormatStyle::LK_JavaScript) {
2615  if (Left.is(TT_JsFatArrow))
2616  return true;
2617  // for await ( ...
2618  if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous &&
2619  Left.Previous->is(tok::kw_for))
2620  return true;
2621  if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) &&
2622  Right.MatchingParen) {
2623  const FormatToken *Next = Right.MatchingParen->getNextNonComment();
2624  // An async arrow function, for example: `x = async () => foo();`,
2625  // as opposed to calling a function called async: `x = async();`
2626  if (Next && Next->is(TT_JsFatArrow))
2627  return true;
2628  }
2629  if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2630  (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2631  return false;
2632  // In tagged template literals ("html`bar baz`"), there is no space between
2633  // the tag identifier and the template string. getIdentifierInfo makes sure
2634  // that the identifier is not a pseudo keyword like `yield`, either.
2635  if (Left.is(tok::identifier) && Keywords.IsJavaScriptIdentifier(Left) &&
2636  Right.is(TT_TemplateString))
2637  return false;
2638  if (Right.is(tok::star) &&
2639  Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2640  return false;
2641  if (Right.isOneOf(tok::l_brace, tok::l_square) &&
2642  Left.isOneOf(Keywords.kw_function, Keywords.kw_yield,
2643  Keywords.kw_extends, Keywords.kw_implements))
2644  return true;
2645  if (Right.is(tok::l_paren)) {
2646  // JS methods can use some keywords as names (e.g. `delete()`).
2647  if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo())
2648  return false;
2649  // Valid JS method names can include keywords, e.g. `foo.delete()` or
2650  // `bar.instanceof()`. Recognize call positions by preceding period.
2651  if (Left.Previous && Left.Previous->is(tok::period) &&
2652  Left.Tok.getIdentifierInfo())
2653  return false;
2654  // Additional unary JavaScript operators that need a space after.
2655  if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof,
2656  tok::kw_void))
2657  return true;
2658  }
2659  if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
2660  tok::kw_const) ||
2661  // "of" is only a keyword if it appears after another identifier
2662  // (e.g. as "const x of y" in a for loop), or after a destructuring
2663  // operation (const [x, y] of z, const {a, b} of c).
2664  (Left.is(Keywords.kw_of) && Left.Previous &&
2665  (Left.Previous->Tok.is(tok::identifier) ||
2666  Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) &&
2667  (!Left.Previous || !Left.Previous->is(tok::period)))
2668  return true;
2669  if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
2670  Left.Previous->is(tok::period) && Right.is(tok::l_paren))
2671  return false;
2672  if (Left.is(Keywords.kw_as) &&
2673  Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
2674  return true;
2675  if (Left.is(tok::kw_default) && Left.Previous &&
2676  Left.Previous->is(tok::kw_export))
2677  return true;
2678  if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2679  return true;
2680  if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2681  return false;
2682  if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
2683  return false;
2684  if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2685  Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2686  return false;
2687  if (Left.is(tok::ellipsis))
2688  return false;
2689  if (Left.is(TT_TemplateCloser) &&
2690  !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2691  Keywords.kw_implements, Keywords.kw_extends))
2692  // Type assertions ('<type>expr') are not followed by whitespace. Other
2693  // locations that should have whitespace following are identified by the
2694  // above set of follower tokens.
2695  return false;
2696  if (Right.is(TT_JsNonNullAssertion))
2697  return false;
2698  if (Left.is(TT_JsNonNullAssertion) &&
2699  Right.isOneOf(Keywords.kw_as, Keywords.kw_in))
2700  return true; // "x! as string", "x! in y"
2701  } else if (Style.Language == FormatStyle::LK_Java) {
2702  if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2703  return true;
2704  if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2706  if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2707  tok::kw_protected) ||
2708  Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2709  Keywords.kw_native)) &&
2710  Right.is(TT_TemplateOpener))
2711  return true;
2712  }
2713  if (Left.is(TT_ImplicitStringLiteral))
2714  return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2715  if (Line.Type == LT_ObjCMethodDecl) {
2716  if (Left.is(TT_ObjCMethodSpecifier))
2717  return true;
2718  if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right))
2719  // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a
2720  // keyword in Objective-C, and '+ (instancetype)new;' is a standard class
2721  // method declaration.
2722  return false;
2723  }
2724  if (Line.Type == LT_ObjCProperty &&
2725  (Right.is(tok::equal) || Left.is(tok::equal)))
2726  return false;
2727 
2728  if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2729  Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2730  return true;
2731  if (Right.is(TT_OverloadedOperatorLParen))
2733  if (Left.is(tok::comma))
2734  return true;
2735  if (Right.is(tok::comma))
2736  return false;
2737  if (Right.is(TT_ObjCBlockLParen))
2738  return true;
2739  if (Right.is(TT_CtorInitializerColon))
2741  if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon)
2742  return false;
2743  if (Right.is(TT_RangeBasedForLoopColon) &&
2745  return false;
2746  if (Right.is(tok::colon)) {
2747  if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2748  !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2749  return false;
2750  if (Right.is(TT_ObjCMethodExpr))
2751  return false;
2752  if (Left.is(tok::question))
2753  return false;
2754  if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2755  return false;
2756  if (Right.is(TT_DictLiteral))
2758  if (Right.is(TT_AttributeColon))
2759  return false;
2760  return true;
2761  }
2762  if (Left.is(TT_UnaryOperator))
2763  return Right.is(TT_BinaryOperator);
2764 
2765  // If the next token is a binary operator or a selector name, we have
2766  // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2767  if (Left.is(TT_CastRParen))
2768  return Style.SpaceAfterCStyleCast ||
2769  Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2770 
2771  if (Left.is(tok::greater) && Right.is(tok::greater)) {
2773  (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral)))
2774  return !Style.Cpp11BracedListStyle;
2775  return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2777  }
2778  if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) ||
2779  Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2780  (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod)))
2781  return false;
2783  Right.getPrecedence() == prec::Assignment)
2784  return false;
2785  if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) &&
2786  (Left.is(tok::identifier) || Left.is(tok::kw_this)))
2787  return false;
2788  if (Right.is(tok::coloncolon) && Left.is(tok::identifier))
2789  // Generally don't remove existing spaces between an identifier and "::".
2790  // The identifier might actually be a macro name such as ALWAYS_INLINE. If
2791  // this turns out to be too lenient, add analysis of the identifier itself.
2792  return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2793  if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment))
2794  return (Left.is(TT_TemplateOpener) &&
2796  !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square,
2797  tok::kw___super, TT_TemplateCloser,
2798  TT_TemplateOpener)) ||
2799  (Left.is(tok ::l_paren) && Style.SpacesInParentheses);
2800  if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2801  return Style.SpacesInAngles;
2802  // Space before TT_StructuredBindingLSquare.
2803  if (Right.is(TT_StructuredBindingLSquare))
2804  return !Left.isOneOf(tok::amp, tok::ampamp) ||
2806  // Space before & or && following a TT_StructuredBindingLSquare.
2807  if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) &&
2808  Right.isOneOf(tok::amp, tok::ampamp))
2810  if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2811  (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2812  !Right.is(tok::r_paren)))
2813  return true;
2814  if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2815  Right.isNot(TT_FunctionTypeLParen))
2817  if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2818  Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2819  return false;
2820  if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2821  Line.startsWith(tok::hash))
2822  return true;
2823  if (Right.is(TT_TrailingUnaryOperator))
2824  return false;
2825  if (Left.is(TT_RegexLiteral))
2826  return false;
2827  return spaceRequiredBetween(Line, Left, Right);
2828 }
2829 
2830 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2831 static bool isAllmanBrace(const FormatToken &Tok) {
2832  return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2833  !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2834 }
2835 
2836 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2837  const FormatToken &Right) {
2838  const FormatToken &Left = *Right.Previous;
2839  if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2840  return true;
2841 
2843  // FIXME: This might apply to other languages and token kinds.
2844  if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
2845  Left.Previous->is(tok::string_literal))
2846  return true;
2847  if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2848  Left.Previous && Left.Previous->is(tok::equal) &&
2849  Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2850  tok::kw_const) &&
2851  // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2852  // above.
2853  !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2854  // Object literals on the top level of a file are treated as "enum-style".
2855  // Each key/value pair is put on a separate line, instead of bin-packing.
2856  return true;
2857  if (Left.is(tok::l_brace) && Line.Level == 0 &&
2858  (Line.startsWith(tok::kw_enum) ||
2859  Line.startsWith(tok::kw_const, tok::kw_enum) ||
2860  Line.startsWith(tok::kw_export, tok::kw_enum) ||
2861  Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum)))
2862  // JavaScript top-level enum key/value pairs are put on separate lines
2863  // instead of bin-packing.
2864  return true;
2865  if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2866  !Left.Children.empty())
2867  // Support AllowShortFunctionsOnASingleLine for JavaScript.
2870  (Left.NestingLevel == 0 && Line.Level == 0 &&
2873  } else if (Style.Language == FormatStyle::LK_Java) {
2874  if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2875  Right.Next->is(tok::string_literal))
2876  return true;
2877  } else if (Style.Language == FormatStyle::LK_Cpp ||
2882  if (Left.isStringLiteral() && Right.isStringLiteral())
2883  return true;
2884  }
2885 
2886  // If the last token before a '}', ']', or ')' is a comma or a trailing
2887  // comment, the intention is to insert a line break after it in order to make
2888  // shuffling around entries easier. Import statements, especially in
2889  // JavaScript, can be an exception to this rule.
2891  const FormatToken *BeforeClosingBrace = nullptr;
2892  if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
2894  Left.is(tok::l_paren))) &&
2895  Left.BlockKind != BK_Block && Left.MatchingParen)
2896  BeforeClosingBrace = Left.MatchingParen->Previous;
2897  else if (Right.MatchingParen &&
2898  (Right.MatchingParen->isOneOf(tok::l_brace,
2899  TT_ArrayInitializerLSquare) ||
2901  Right.MatchingParen->is(tok::l_paren))))
2902  BeforeClosingBrace = &Left;
2903  if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2904  BeforeClosingBrace->isTrailingComment()))
2905  return true;
2906  }
2907 
2908  if (Right.is(tok::comment))
2909  return Left.BlockKind != BK_BracedInit &&
2910  Left.isNot(TT_CtorInitializerColon) &&
2911  (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2912  if (Left.isTrailingComment())
2913  return true;
2914  if (Right.Previous->IsUnterminatedLiteral)
2915  return true;
2916  if (Right.is(tok::lessless) && Right.Next &&
2917  Right.Previous->is(tok::string_literal) &&
2918  Right.Next->is(tok::string_literal))
2919  return true;
2920  if (Right.Previous->ClosesTemplateDeclaration &&
2921  Right.Previous->MatchingParen &&
2922  Right.Previous->MatchingParen->NestingLevel == 0 &&
2924  return true;
2925  if (Right.is(TT_CtorInitializerComma) &&
2928  return true;
2929  if (Right.is(TT_CtorInitializerColon) &&
2932  return true;
2933  // Break only if we have multiple inheritance.
2935  Right.is(TT_InheritanceComma))
2936  return true;
2937  if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2938  // Multiline raw string literals are special wrt. line breaks. The author
2939  // has made a deliberate choice and might have aligned the contents of the
2940  // string literal accordingly. Thus, we try keep existing line breaks.
2941  return Right.IsMultiline && Right.NewlinesBefore > 0;
2942  if ((Right.Previous->is(tok::l_brace) ||
2943  (Right.Previous->is(tok::less) && Right.Previous->Previous &&
2944  Right.Previous->Previous->is(tok::equal))) &&
2945  Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) {
2946  // Don't put enums or option definitions onto single lines in protocol
2947  // buffers.
2948  return true;
2949  }
2950  if (Right.is(TT_InlineASMBrace))
2951  return Right.HasUnescapedNewline;
2952  if (isAllmanBrace(Left) || isAllmanBrace(Right))
2953  return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2954  (Line.startsWith(tok::kw_typedef, tok::kw_enum) &&
2956  (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2957  (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2958  if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2959  return true;
2960 
2963  Left.is(TT_LeadingJavaAnnotation) &&
2964  Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2965  (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2966  return true;
2967 
2968  if (Right.is(TT_ProtoExtensionLSquare))
2969  return true;
2970 
2971  // In text proto instances if a submessage contains at least 2 entries and at
2972  // least one of them is a submessage, like A { ... B { ... } ... },
2973  // put all of the entries of A on separate lines by forcing the selector of
2974  // the submessage B to be put on a newline.
2975  //
2976  // Example: these can stay on one line:
2977  // a { scalar_1: 1 scalar_2: 2 }
2978  // a { b { key: value } }
2979  //
2980  // and these entries need to be on a new line even if putting them all in one
2981  // line is under the column limit:
2982  // a {
2983  // scalar: 1
2984  // b { key: value }
2985  // }
2986  //
2987  // We enforce this by breaking before a submessage field that has previous
2988  // siblings, *and* breaking before a field that follows a submessage field.
2989  //
2990  // Be careful to exclude the case [proto.ext] { ... } since the `]` is
2991  // the TT_SelectorName there, but we don't want to break inside the brackets.
2992  //
2993  // Another edge case is @submessage { key: value }, which is a common
2994  // substitution placeholder. In this case we want to keep `@` and `submessage`
2995  // together.
2996  //
2997  // We ensure elsewhere that extensions are always on their own line.
3000  Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) {
3001  // Keep `@submessage` together in:
3002  // @submessage { key: value }
3003  if (Right.Previous && Right.Previous->is(tok::at))
3004  return false;
3005  // Look for the scope opener after selector in cases like:
3006  // selector { ...
3007  // selector: { ...
3008  // selector: @base { ...
3009  FormatToken *LBrace = Right.Next;
3010  if (LBrace && LBrace->is(tok::colon)) {
3011  LBrace = LBrace->Next;
3012  if (LBrace && LBrace->is(tok::at)) {
3013  LBrace = LBrace->Next;
3014  if (LBrace)
3015  LBrace = LBrace->Next;
3016  }
3017  }
3018  if (LBrace &&
3019  // The scope opener is one of {, [, <:
3020  // selector { ... }
3021  // selector [ ... ]
3022  // selector < ... >
3023  //
3024  // In case of selector { ... }, the l_brace is TT_DictLiteral.
3025  // In case of an empty selector {}, the l_brace is not TT_DictLiteral,
3026  // so we check for immediately following r_brace.
3027  ((LBrace->is(tok::l_brace) &&
3028  (LBrace->is(TT_DictLiteral) ||
3029  (LBrace->Next && LBrace->Next->is(tok::r_brace)))) ||
3030  LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) {
3031  // If Left.ParameterCount is 0, then this submessage entry is not the
3032  // first in its parent submessage, and we want to break before this entry.
3033  // If Left.ParameterCount is greater than 0, then its parent submessage
3034  // might contain 1 or more entries and we want to break before this entry
3035  // if it contains at least 2 entries. We deal with this case later by
3036  // detecting and breaking before the next entry in the parent submessage.
3037  if (Left.ParameterCount == 0)
3038  return true;
3039  // However, if this submessage is the first entry in its parent
3040  // submessage, Left.ParameterCount might be 1 in some cases.
3041  // We deal with this case later by detecting an entry
3042  // following a closing paren of this submessage.
3043  }
3044 
3045  // If this is an entry immediately following a submessage, it will be
3046  // preceded by a closing paren of that submessage, like in:
3047  // left---. .---right
3048  // v v
3049  // sub: { ... } key: value
3050  // If there was a comment between `}` an `key` above, then `key` would be
3051  // put on a new line anyways.
3052  if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square))
3053  return true;
3054  }
3055 
3056  // Deal with lambda arguments in C++ - we want consistent line breaks whether
3057  // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced
3058  // as aggressive line breaks are placed when the lambda is not the last arg.
3061  Left.is(tok::l_paren) && Left.BlockParameterCount > 0 &&
3062  !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) {
3063  // Multiple lambdas in the same function call force line breaks.
3064  if (Left.BlockParameterCount > 1)
3065  return true;
3066 
3067  // A lambda followed by another arg forces a line break.
3068  if (!Left.Role)
3069  return false;
3070  auto Comma = Left.Role->lastComma();
3071  if (!Comma)
3072  return false;
3073  auto Next = Comma->getNextNonComment();
3074  if (!Next)
3075  return false;
3076  if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret))
3077  return true;
3078  }
3079 
3080  return false;
3081 }
3082 
3083 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
3084  const FormatToken &Right) {
3085  const FormatToken &Left = *Right.Previous;
3086 
3087  // Language-specific stuff.
3089  if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
3090  Keywords.kw_implements))
3091  return false;
3092  if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
3093  Keywords.kw_implements))
3094  return true;
3095  } else if (Style.Language == FormatStyle::LK_JavaScript) {
3096  const FormatToken *NonComment = Right.getPreviousNonComment();
3097  if (NonComment &&
3098  NonComment->isOneOf(
3099  tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break,
3100  tok::kw_throw, Keywords.kw_interface, Keywords.kw_type,
3101  tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected,
3102  Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get,
3103  Keywords.kw_set, Keywords.kw_async, Keywords.kw_await))
3104  return false; // Otherwise automatic semicolon insertion would trigger.
3105  if (Right.NestingLevel == 0 &&
3106  (Left.Tok.getIdentifierInfo() ||
3107  Left.isOneOf(tok::r_square, tok::r_paren)) &&
3108  Right.isOneOf(tok::l_square, tok::l_paren))
3109  return false; // Otherwise automatic semicolon insertion would trigger.
3110  if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
3111  return false;
3112  if (Left.is(TT_JsTypeColon))
3113  return true;
3114  // Don't wrap between ":" and "!" of a strict prop init ("field!: type;").
3115  if (Left.is(tok::exclaim) && Right.is(tok::colon))
3116  return false;
3117  // Look for is type annotations like:
3118  // function f(): a is B { ... }
3119  // Do not break before is in these cases.
3120  if (Right.is(Keywords.kw_is)) {
3121  const FormatToken* Next = Right.getNextNonComment();
3122  // If `is` is followed by a colon, it's likely that it's a dict key, so
3123  // ignore it for this check.
3124  // For example this is common in Polymer:
3125  // Polymer({
3126  // is: 'name',
3127  // ...
3128  // });
3129  if (!Next || !Next->is(tok::colon))
3130  return false;
3131  }
3132  if (Left.is(Keywords.kw_in))
3134  if (Right.is(Keywords.kw_in))
3136  if (Right.is(Keywords.kw_as))
3137  return false; // must not break before as in 'x as type' casts
3138  if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) {
3139  // extends and infer can appear as keywords in conditional types:
3140  // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types
3141  // do not break before them, as the expressions are subject to ASI.
3142  return false;
3143  }
3144  if (Left.is(Keywords.kw_as))
3145  return true;
3146  if (Left.is(TT_JsNonNullAssertion))
3147  return true;
3148  if (Left.is(Keywords.kw_declare) &&
3149  Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
3150  Keywords.kw_function, tok::kw_class, tok::kw_enum,
3151  Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
3152  Keywords.kw_let, tok::kw_const))
3153  // See grammar for 'declare' statements at:
3154  // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
3155  return false;
3156  if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
3157  Right.isOneOf(tok::identifier, tok::string_literal))
3158  return false; // must not break in "module foo { ...}"
3159  if (Right.is(TT_TemplateString) && Right.closesScope())
3160  return false;
3161  if (Left.is(TT_TemplateString) && Left.opensScope())
3162  return true;
3163  }
3164 
3165  if (Left.is(tok::at))
3166  return false;
3167  if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
3168  return false;
3169  if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
3170  return !Right.is(tok::l_paren);
3171  if (Right.is(TT_PointerOrReference))
3172  return Line.IsMultiVariableDeclStmt ||
3174  (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
3175  if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
3176  Right.is(tok::kw_operator))
3177  return true;
3178  if (Left.is(TT_PointerOrReference))
3179  return false;
3180  if (Right.isTrailingComment())
3181  // We rely on MustBreakBefore being set correctly here as we should not
3182  // change the "binding" behavior of a comment.
3183  // The first comment in a braced lists is always interpreted as belonging to
3184  // the first list element. Otherwise, it should be placed outside of the
3185  // list.
3186  return Left.BlockKind == BK_BracedInit ||
3187  (Left.is(TT_CtorInitializerColon) &&
3189  if (Left.is(tok::question) && Right.is(tok::colon))
3190  return false;
3191  if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
3193  if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
3195  if (Left.is(TT_InheritanceColon))
3197  if (Right.is(TT_InheritanceColon))
3199  if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
3200  Left.isNot(TT_SelectorName))
3201  return true;
3202 
3203  if (Right.is(tok::colon) &&
3204  !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
3205  return false;
3206  if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) {
3210  return false;
3211  // Prevent cases like:
3212  //
3213  // submessage:
3214  // { key: valueeeeeeeeeeee }
3215  //
3216  // when the snippet does not fit into one line.
3217  // Prefer:
3218  //
3219  // submessage: {
3220  // key: valueeeeeeeeeeee
3221  // }
3222  //
3223  // instead, even if it is longer by one line.
3224  //
3225  // Note that this allows allows the "{" to go over the column limit
3226  // when the column limit is just between ":" and "{", but that does
3227  // not happen too often and alternative formattings in this case are
3228  // not much better.
3229  //
3230  // The code covers the cases:
3231  //
3232  // submessage: { ... }
3233  // submessage: < ... >
3234  // repeated: [ ... ]
3235  if (((Right.is(tok::l_brace) || Right.is(tok::less)) &&
3236  Right.is(TT_DictLiteral)) ||
3237  Right.is(TT_ArrayInitializerLSquare))
3238  return false;
3239  }
3240  return true;
3241  }
3242  if (Right.is(tok::r_square) && Right.MatchingParen &&
3243  Right.MatchingParen->is(TT_ProtoExtensionLSquare))
3244  return false;
3245  if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
3246  Right.Next->is(TT_ObjCMethodExpr)))
3247  return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
3248  if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
3249  return true;
3250  if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
3251  return true;
3252  if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
3253  TT_OverloadedOperator))
3254  return false;
3255  if (Left.is(TT_RangeBasedForLoopColon))
3256  return true;
3257  if (Right.is(TT_RangeBasedForLoopColon))
3258  return false;
3259  if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
3260  return true;
3261  if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
3262  Left.is(tok::kw_operator))
3263  return false;
3264  if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
3265  Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
3266  return false;
3267  if (Left.is(tok::equal) && Right.is(tok::l_brace) &&
3269  return false;
3270  if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
3271  return false;
3272  if (Left.is(tok::l_paren) && Left.Previous &&
3273  (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
3274  return false;
3275  if (Right.is(TT_ImplicitStringLiteral))
3276  return false;
3277 
3278  if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
3279  return false;
3280  if (Right.is(tok::r_square) && Right.MatchingParen &&
3281  Right.MatchingParen->is(TT_LambdaLSquare))
3282  return false;
3283 
3284  // We only break before r_brace if there was a corresponding break before
3285  // the l_brace, which is tracked by BreakBeforeClosingBrace.
3286  if (Right.is(tok::r_brace))
3287  return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
3288 
3289  // Allow breaking after a trailing annotation, e.g. after a method
3290  // declaration.
3291  if (Left.is(TT_TrailingAnnotation))
3292  return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
3293  tok::less, tok::coloncolon);
3294 
3295  if (Right.is(tok::kw___attribute) ||
3296  (Right.is(tok::l_square) && Right.is(TT_AttributeSquare)))
3297  return true;
3298 
3299  if (Left.is(tok::identifier) && Right.is(tok::string_literal))
3300  return true;
3301 
3302  if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
3303  return true;
3304 
3305  if (Left.is(TT_CtorInitializerColon))
3307  if (Right.is(TT_CtorInitializerColon))
3309  if (Left.is(TT_CtorInitializerComma) &&
3311  return false;
3312  if (Right.is(TT_CtorInitializerComma) &&
3314  return true;
3315  if (Left.is(TT_InheritanceComma) &&
3317  return false;
3318  if (Right.is(TT_InheritanceComma) &&
3320  return true;
3321  if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
3322  (Left.is(tok::less) && Right.is(tok::less)))
3323  return false;
3324  if (Right.is(TT_BinaryOperator) &&
3327  Right.getPrecedence() != prec::Assignment))
3328  return true;
3329  if (Left.is(TT_ArrayInitializerLSquare))
3330  return true;
3331  if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
3332  return true;
3333  if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
3334  !Left.isOneOf(tok::arrowstar, tok::lessless) &&
3337  Left.getPrecedence() == prec::Assignment))
3338  return true;
3339  if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) ||
3340  (Left.is(tok::r_square) && Right.is(TT_AttributeSquare)))
3341  return false;
3342  return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
3343  tok::kw_class, tok::kw_struct, tok::comment) ||
3344  Right.isMemberAccess() ||
3345  Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
3346  tok::colon, tok::l_square, tok::at) ||
3347  (Left.is(tok::r_paren) &&
3348  Right.isOneOf(tok::identifier, tok::kw_const)) ||
3349  (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) ||
3350  (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser));
3351 }
3352 
3353 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
3354  llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n";
3355  const FormatToken *Tok = Line.First;
3356  while (Tok) {
3357  llvm::errs() << " M=" << Tok->MustBreakBefore
3358  << " C=" << Tok->CanBreakBefore
3359  << " T=" << getTokenTypeName(Tok->Type)
3360  << " S=" << Tok->SpacesRequiredBefore
3361  << " B=" << Tok->BlockParameterCount
3362  << " BK=" << Tok->BlockKind << " P=" << Tok->SplitPenalty
3363  << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength
3364  << " PPK=" << Tok->PackingKind << " FakeLParens=";
3365  for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
3366  llvm::errs() << Tok->FakeLParens[i] << "/";
3367  llvm::errs() << " FakeRParens=" << Tok->FakeRParens;
3368  llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo();
3369  llvm::errs() << " Text='" << Tok->TokenText << "'\n";
3370  if (!Tok->Next)
3371  assert(Tok == Line.Last);
3372  Tok = Tok->Next;
3373  }
3374  llvm::errs() << "----\n";
3375 }
3376 
3377 } // namespace format
3378 } // namespace clang
Always break after template declaration.
Definition: Format.h:389
Always break after the return type of top-level definitions.
Definition: Format.h:332
bool endsSequence(A K1, Ts... Tokens) const
true if this token ends a sequence with the given tokens in order, following the Previous pointers...
Definition: FormatToken.h:343
if(T->getSizeExpr()) TRY_TO(TraverseStmt(T -> getSizeExpr()))
unsigned NestingLevel
The nesting level of this token, i.e.
Definition: FormatToken.h:228
Break inheritance list after the colon and commas.
Definition: Format.h:921
bool AlwaysBreakBeforeMultilineStrings
If true, always break before multiline string literals.
Definition: Format.h:354
unsigned PenaltyBreakBeforeFirstCallParameter
The penalty for breaking a function call after call(.
Definition: Format.h:1378
Token Tok
The Token.
Definition: FormatToken.h:128
Should be used for Protocol Buffers (https://developers.google.com/protocol-buffers/).
Definition: Format.h:1231
bool is(tok::TokenKind K) const
is/isNot - Predicates to check if this token is a specific kind, as in "if (Tok.is(tok::l_brace)) {...
Definition: Token.h:95
Defines the SourceManager interface.
bool SpaceBeforeRangeBasedForLoopColon
If false, spaces will be removed before range-based for loop colon.
Definition: Format.h:1610
std::unique_ptr< TokenRole > Role
A token can have a special role that can carry extra information about the token&#39;s formatting...
Definition: FormatToken.h:204
Break constructor initializers before the colon and commas, and align the commas with the colon...
Definition: Format.h:854
unsigned OriginalColumn
The original 0-based column of this token, including expanded tabs.
Definition: FormatToken.h:215
static LLVM_ATTRIBUTE_UNUSED void printDebugInfo(const UnwrappedLine &Line, StringRef Prefix="")
bool isMemberAccess() const
Returns true if this is a "." or "->" accessing a member.
Definition: FormatToken.h:391
bool isFunctionLikeKeyword() const
Returns true if this is a keyword that can be used like a function call (e.g.
Definition: FormatToken.h:425
const FormatToken * getNextNonComment() const
Returns the next token ignoring comments.
Definition: FormatToken.h:481
Parser - This implements a parser for the C family of languages.
Definition: Parser.h:57
bool SpaceAfterTemplateKeyword
If true, a space will be inserted after the &#39;template&#39; keyword.
Definition: Format.h:1530
PointerAlignmentStyle PointerAlignment
Pointer and reference alignment style.
Definition: Format.h:1419
Align pointer to the left.
Definition: Format.h:1405
Should be used for C, C++.
Definition: Format.h:1222
bool IsMultiline
Whether the token text contains newlines (escaped or not).
Definition: FormatToken.h:157
bool AfterEnum
Wrap enum definitions.
Definition: Format.h:663
unsigned TotalLength
The total length of the unwrapped line up to and including this token.
Definition: FormatToken.h:211
bool isBinaryOperator() const
Definition: FormatToken.h:413
unsigned NewlinesBefore
The number of newlines immediately before the Token.
Definition: FormatToken.h:134
FormatToken * Next
The next token in the unwrapped line.
Definition: FormatToken.h:293
Break after operators.
Definition: Format.h:451
bool isLiteral() const
Return true if this is a "literal", like a numeric constant, string, etc.
Definition: Token.h:113
bool CanBeExpression
tok::TokenKind ContextKind
unsigned UnbreakableTailLength
The length of following tokens until the next natural split point, or the next token that can be brok...
Definition: FormatToken.h:219
bool closesScope() const
Returns whether Tok is )]} or a closing > of a template or in protos.
Definition: FormatToken.h:381
bool JavaScriptWrapImports
Whether to wrap JavaScript import/export statements.
Definition: Format.h:1201
Always break after the return type.
Definition: Format.h:292
unsigned SplitPenalty
Penalty for inserting a line break before this token.
Definition: FormatToken.h:234
bool ColonIsForRangeExpr
prec::Level getPrecedence() const
Definition: FormatToken.h:467
unsigned ParameterCount
Number of parameters, if this is "(", "[" or "<".
Definition: FormatToken.h:192
unsigned FakeRParens
Insert this many fake ) after this token for correct indentation.
Definition: FormatToken.h:260
bool SpaceInEmptyParentheses
If true, spaces may be inserted into ().
Definition: Format.h:1622
bool CanBreakBefore
true if it is allowed to break before this token.
Definition: FormatToken.h:186
Should be used for Java.
Definition: Format.h:1224
FormatToken * Previous
The previous token in the unwrapped line.
Definition: FormatToken.h:290
Always break after the return type of function definitions.
Definition: Format.h:320
Only merge functions defined inside a class.
Definition: Format.h:211
This file implements a token annotator, i.e.
bool BreakAfterJavaFieldAnnotations
Break after each annotation on a field in Java files.
Definition: Format.h:874
bool ConstructorInitializerAllOnOneLineOrOnePerLine
If the constructor initializers don&#39;t fit on a line, put each initializer on its own line...
Definition: Format.h:966
unsigned OperatorIndex
If this is an operator (or "."/"->") in a sequence of operators with the same precedence, contains the 0-based operator index.
Definition: FormatToken.h:270
unsigned SpacesRequiredBefore
The number of spaces that should be inserted before this token.
Definition: FormatToken.h:183
bool isNot(T Kind) const
Definition: FormatToken.h:323
bool SpacesInParentheses
If true, spaces will be inserted after ( and before ).
Definition: Format.h:1670
const FormatToken & Tok
ReturnTypeBreakingStyle AlwaysBreakAfterReturnType
The function declaration return type breaking style to use.
Definition: Format.h:340
unsigned BlockParameterCount
Number of parameters that are nested blocks, if this is "(", "[" or "<".
Definition: FormatToken.h:196
bool InInheritanceList
bool SpaceBeforeAssignmentOperators
If false, spaces will be removed before assignment operators.
Definition: Format.h:1538
SpaceBeforeParensOptions SpaceBeforeParens
Defines in which cases to put a space before opening parentheses.
Definition: Format.h:1602
bool SpaceBeforeCtorInitializerColon
If false, spaces will be removed before constructor initializer colon.
Definition: Format.h:1557
unsigned PenaltyBreakAssignment
The penalty for breaking around an assignment operator.
Definition: Format.h:1375
const char * getName() const
Definition: Token.h:166
Always put a space before opening parentheses, except when it&#39;s prohibited by the syntax rules (in fu...
Definition: Format.h:1598
bool mightBeFunctionDefinition() const
true if this line looks like a function definition instead of a function declaration.
unsigned ColumnLimit
The column limit.
Definition: Format.h:884
Never merge functions into a single line.
Definition: Format.h:197
FormatToken * getPreviousNonComment() const
Returns the previous token ignoring comments.
Definition: FormatToken.h:473
AnnotatingParser & P
BracketAlignmentStyle AlignAfterOpenBracket
If true, horizontally aligns arguments after an open bracket.
Definition: Format.h:81
NodeId Parent
Definition: ASTDiff.cpp:192
bool isLabelString() const
Returns true if this is a string literal that&#39;s like a label, e.g.
Definition: FormatToken.h:445
bool isOneOf(A K1, B K2) const
Definition: FormatToken.h:316
Break after return type automatically.
Definition: Format.h:276
Only merge empty functions.
Definition: Format.h:219
Should be used for JavaScript.
Definition: Format.h:1226
const AnnotatedLine * Line
StateNode * Previous
SmallVector< AnnotatedLine *, 0 > Children
ParameterPackingKind PackingKind
If this is an opening parenthesis, how are the parameters packed?
Definition: FormatToken.h:207
bool SpacesInContainerLiterals
If true, spaces are inserted inside container literals (e.g.
Definition: Format.h:1656
BraceWrappingFlags BraceWrapping
Control of individual brace wrapping cases.
Definition: Format.h:822
bool SpacesInAngles
If true, spaces will be inserted after < and before > in template argument lists. ...
Definition: Format.h:1647
A wrapper around a Token storing information about the whitespace characters preceding it...
Definition: FormatToken.h:124
void setCommentLineLevels(SmallVectorImpl< AnnotatedLine *> &Lines)
Adapts the indent levels of comment lines to the indent of the subsequent line.
static unsigned maxNestingDepth(const AnnotatedLine &Line)
FormatToken * FirstStartOfName
SourceLocation getEnd() const
bool isTrailingComment() const
Definition: FormatToken.h:418
void annotate(AnnotatedLine &Line)
Don&#39;t align, instead use ContinuationIndentWidth, e.g.
Definition: Format.h:67
const char * getTokenTypeName(TokenType Type)
Determines the name of a token type.
Definition: FormatToken.cpp:25
static bool isAllmanBrace(const FormatToken &Tok)
bool is(tok::TokenKind Kind) const
Definition: FormatToken.h:307
bool InCpp11AttributeSpecifier
IdentifierInfo * getIdentifierInfo() const
Definition: Token.h:177
Break constructor initializers after the colon and commas.
Definition: Format.h:861
bool BreakBeforeTernaryOperators
If true, ternary operators will be placed after line breaks.
Definition: Format.h:836
SourceRange WhitespaceRange
The range of the whitespace immediately preceding the Token.
Definition: FormatToken.h:141
TokenKind
Provides a simple uniform namespace for tokens from all C languages.
Definition: TokenKinds.h:25
bool SpacesInCStyleCastParentheses
If true, spaces may be inserted into C style casts.
Definition: Format.h:1663
bool IsUnterminatedLiteral
Set to true if this token is an unterminated literal.
Definition: FormatToken.h:175
bool startsWith(Ts... Tokens) const
true if this line starts with the given tokens in order, ignoring comments.
bool SpacesInSquareBrackets
If true, spaces will be inserted after [ and before ].
Definition: Format.h:1679
StringRef TokenText
The raw text of the token.
Definition: FormatToken.h:172
SmallVector< prec::Level, 4 > FakeLParens
Stores the number of required fake parentheses and the corresponding operator precedence.
Definition: FormatToken.h:258
tok::ObjCKeywordKind getObjCKeywordID() const
Return the ObjC keyword kind.
Definition: Lexer.cpp:67
BreakTemplateDeclarationsStyle AlwaysBreakTemplateDeclarations
The template declaration breaking style to use.
Definition: Format.h:393
bool isSimpleTypeSpecifier() const
Determine whether the token is a simple-type-specifier.
Definition: FormatToken.cpp:40
The FormatStyle is used to configure the formatting to follow specific guidelines.
Definition: Format.h:50
Never put a space before opening parentheses.
Definition: Format.h:1576
unsigned PenaltyReturnTypeOnItsOwnLine
Penalty for putting the return type of a function onto its own line.
Definition: Format.h:1397
unsigned IndentLevel
The indent level of this token. Copied from the surrounding line.
Definition: FormatToken.h:231
LanguageKind Language
Language, this format style is targeted at.
Definition: Format.h:1241
Optional< types::ID > Type
bool ColonIsObjCMethodExpr
Dataflow Directional Tag Classes.
ShortFunctionStyle AllowShortFunctionsOnASingleLine
Dependent on the value, int f() { return 0; } can be put on a single line.
Definition: Format.h:243
unsigned ColumnWidth
The width of the non-whitespace parts of the token (or its first line for multi-line tokens) in colum...
Definition: FormatToken.h:150
BreakInheritanceListStyle BreakInheritanceList
The inheritance list style to use.
Definition: Format.h:925
Use C++03-compatible syntax.
Definition: Format.h:1684
Always break after the return types of top-level functions.
Definition: Format.h:305
Should be used for TableGen code.
Definition: Format.h:1233
bool IsExpression
Use features of C++11, C++14 and C++1z (e.g.
Definition: Format.h:1687
bool Cpp11BracedListStyle
If true, format braced lists as best suited for C++11 braced lists.
Definition: Format.h:1002
static bool isFunctionDeclarationName(const FormatToken &Current, const AnnotatedLine &Line)
bool SpaceBeforeCpp11BracedList
If true, a space will be inserted before a C++11 braced list used to initialize an object (after the ...
Definition: Format.h:1549
bool AfterStruct
Wrap struct definitions.
Definition: Format.h:713
Break inheritance list before the colon and commas, and align the commas with the colon...
Definition: Format.h:913
void calculateFormattingInformation(AnnotatedLine &Line)
bool SpaceAfterCStyleCast
If true, a space is inserted after C style casts.
Definition: Format.h:1523
Should be used for Objective-C, Objective-C++.
Definition: Format.h:1228
LanguageStandard Standard
Format compatible with this standard, e.g.
Definition: Format.h:1694
FormatToken * NextOperator
If this is an operator (or "."/"->") in a sequence of operators with the same precedence, points to the next operator.
Definition: FormatToken.h:274
BreakConstructorInitializersStyle BreakConstructorInitializers
The constructor initializers style to use.
Definition: Format.h:865
unsigned BindingStrength
bool ClosesTemplateDeclaration
true if this is the ">" of "template<..>".
Definition: FormatToken.h:189
FormatToken * MatchingParen
If this is a bracket, this points to the matching one.
Definition: FormatToken.h:287
SmallVector< AnnotatedLine *, 1 > Children
If this token starts a block, this contains all the unwrapped lines in it.
Definition: FormatToken.h:297
FormatToken * FirstObjCSelectorName
bool InTemplateArgument
bool ColonIsDictLiteral
The parameter type of a method or function.
BinaryOperatorStyle BreakBeforeBinaryOperators
The way to wrap binary operators.
Definition: Format.h:479
bool opensBlockOrBlockTypeList(const FormatStyle &Style) const
Returns true if this tokens starts a block-type list, i.e.
Definition: FormatToken.h:490
Break before operators.
Definition: Format.h:475
bool ObjCSpaceBeforeProtocolList
Add a space in front of an Objective-C protocol list, i.e.
Definition: Format.h:1372
unsigned SpacesBeforeTrailingComments
The number of spaces before trailing line comments (// - comments).
Definition: Format.h:1638
bool opensScope() const
Returns whether Tok is ([{ or an opening < of a template or in protos.
Definition: FormatToken.h:371
bool SpaceBeforeInheritanceColon
If false, spaces will be removed before inheritance colon.
Definition: Format.h:1564
bool AllowShortBlocksOnASingleLine
Allows contracting simple braced statements to a single line.
Definition: Format.h:178
unsigned LongestObjCSelectorName
bool MustBreakBefore
Whether there must be a line break before this token.
Definition: FormatToken.h:166
unsigned PenaltyBreakTemplateDeclaration
The penalty for breaking after template declaration.
Definition: Format.h:1390
__DEVICE__ int max(int __a, int __b)
bool AfterClass
Wrap class definitions.
Definition: Format.h:633
Should be used for Protocol Buffer messages in text format (https://developers.google.com/protocol-buffers/).
Definition: Format.h:1236
prec::Level getBinOpPrecedence(tok::TokenKind Kind, bool GreaterThanIsOperator, bool CPlusPlus11)
Return the precedence of the specified binary operator token.
Align pointer to the right.
Definition: Format.h:1410
unsigned BindingStrength
The binding strength of a token.
Definition: FormatToken.h:224
bool CaretFound
bool InCtorInitializer
bool isStringLiteral() const
Definition: FormatToken.h:347
SourceLocation getBegin() const
unsigned MaxEmptyLinesToKeep
The maximum number of consecutive empty lines to keep.
Definition: Format.h:1285
bool HasUnescapedNewline
Whether there is at least one unescaped newline before the Token.
Definition: FormatToken.h:138
BraceBlockKind BlockKind
Contains the kind of block if this token is a brace.
Definition: FormatToken.h:178
bool PartOfMultiVariableDeclStmt
Is this token part of a DeclStmt defining multiple variables?
Definition: FormatToken.h:279
unsigned ParameterIndex
The 0-based index of the parameter/argument.
Definition: FormatToken.h:251
bool IsForEachMacro
const FormatStyle & Style