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 /// \brief 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"
17 #include "clang/Basic/SourceManager.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 /// \brief A parser that gathers additional information about tokens.
29 ///
30 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
31 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
32 /// into template parameter lists.
33 class AnnotatingParser {
34 public:
35   AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line,
36                    const AdditionalKeywords &Keywords)
37       : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false),
38         Keywords(Keywords) {
39     Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
40     resetTokenMetadata(CurrentToken);
41   }
42 
43 private:
44   bool parseAngle() {
45     if (!CurrentToken || !CurrentToken->Previous)
46       return false;
47     if (NonTemplateLess.count(CurrentToken->Previous))
48       return false;
49 
50     const FormatToken& Previous = *CurrentToken->Previous;
51     if (Previous.Previous) {
52       if (Previous.Previous->Tok.isLiteral())
53         return false;
54       if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 &&
55           (!Previous.Previous->MatchingParen ||
56            !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen)))
57         return false;
58     }
59 
60     FormatToken *Left = CurrentToken->Previous;
61     Left->ParentBracket = Contexts.back().ContextKind;
62     ScopedContextCreator ContextCreator(*this, tok::less, 12);
63 
64     // If this angle is in the context of an expression, we need to be more
65     // hesitant to detect it as opening template parameters.
66     bool InExprContext = Contexts.back().IsExpression;
67 
68     Contexts.back().IsExpression = false;
69     // If there's a template keyword before the opening angle bracket, this is a
70     // template parameter, not an argument.
71     Contexts.back().InTemplateArgument =
72         Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
73 
74     if (Style.Language == FormatStyle::LK_Java &&
75         CurrentToken->is(tok::question))
76       next();
77 
78     while (CurrentToken) {
79       if (CurrentToken->is(tok::greater)) {
80         Left->MatchingParen = CurrentToken;
81         CurrentToken->MatchingParen = Left;
82         CurrentToken->Type = TT_TemplateCloser;
83         next();
84         return true;
85       }
86       if (CurrentToken->is(tok::question) &&
87           Style.Language == FormatStyle::LK_Java) {
88         next();
89         continue;
90       }
91       if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
92           (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext))
93         return false;
94       // If a && or || is found and interpreted as a binary operator, this set
95       // of angles is likely part of something like "a < b && c > d". If the
96       // angles are inside an expression, the ||/&& might also be a binary
97       // operator that was misinterpreted because we are parsing template
98       // parameters.
99       // FIXME: This is getting out of hand, write a decent parser.
100       if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
101           CurrentToken->Previous->is(TT_BinaryOperator) &&
102           Contexts[Contexts.size() - 2].IsExpression &&
103           !Line.startsWith(tok::kw_template))
104         return false;
105       updateParameterCount(Left, CurrentToken);
106       if (!consumeToken())
107         return false;
108     }
109     return false;
110   }
111 
112   bool parseParens(bool LookForDecls = false) {
113     if (!CurrentToken)
114       return false;
115     FormatToken *Left = CurrentToken->Previous;
116     Left->ParentBracket = Contexts.back().ContextKind;
117     ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
118 
119     // FIXME: This is a bit of a hack. Do better.
120     Contexts.back().ColonIsForRangeExpr =
121         Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
122 
123     bool StartsObjCMethodExpr = false;
124     if (CurrentToken->is(tok::caret)) {
125       // (^ can start a block type.
126       Left->Type = TT_ObjCBlockLParen;
127     } else if (FormatToken *MaybeSel = Left->Previous) {
128       // @selector( starts a selector.
129       if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
130           MaybeSel->Previous->is(tok::at)) {
131         StartsObjCMethodExpr = true;
132       }
133     }
134 
135     if (Left->is(TT_OverloadedOperatorLParen)) {
136       Contexts.back().IsExpression = false;
137     } else if (Left->Previous &&
138         (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype,
139                                  tok::kw_if, tok::kw_while, tok::l_paren,
140                                  tok::comma) ||
141          Left->Previous->is(TT_BinaryOperator))) {
142       // static_assert, if and while usually contain expressions.
143       Contexts.back().IsExpression = true;
144     } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
145                Left->Previous->MatchingParen &&
146                Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
147       // This is a parameter list of a lambda expression.
148       Contexts.back().IsExpression = false;
149     } else if (Line.InPPDirective &&
150                (!Left->Previous || !Left->Previous->is(tok::identifier))) {
151       Contexts.back().IsExpression = true;
152     } else if (Contexts[Contexts.size() - 2].CaretFound) {
153       // This is the parameter list of an ObjC block.
154       Contexts.back().IsExpression = false;
155     } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
156       Left->Type = TT_AttributeParen;
157     } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
158       // The first argument to a foreach macro is a declaration.
159       Contexts.back().IsForEachMacro = true;
160       Contexts.back().IsExpression = false;
161     } else if (Left->Previous && Left->Previous->MatchingParen &&
162                Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
163       Contexts.back().IsExpression = false;
164     } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
165       bool IsForOrCatch =
166           Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
167       Contexts.back().IsExpression = !IsForOrCatch;
168     }
169 
170     if (StartsObjCMethodExpr) {
171       Contexts.back().ColonIsObjCMethodExpr = true;
172       Left->Type = TT_ObjCMethodExpr;
173     }
174 
175     bool MightBeFunctionType = CurrentToken->isOneOf(tok::star, tok::amp) &&
176                                !Contexts[Contexts.size() - 2].IsExpression;
177     bool HasMultipleLines = false;
178     bool HasMultipleParametersOnALine = false;
179     bool MightBeObjCForRangeLoop =
180         Left->Previous && Left->Previous->is(tok::kw_for);
181     while (CurrentToken) {
182       // LookForDecls is set when "if (" has been seen. Check for
183       // 'identifier' '*' 'identifier' followed by not '=' -- this
184       // '*' has to be a binary operator but determineStarAmpUsage() will
185       // categorize it as an unary operator, so set the right type here.
186       if (LookForDecls && CurrentToken->Next) {
187         FormatToken *Prev = CurrentToken->getPreviousNonComment();
188         if (Prev) {
189           FormatToken *PrevPrev = Prev->getPreviousNonComment();
190           FormatToken *Next = CurrentToken->Next;
191           if (PrevPrev && PrevPrev->is(tok::identifier) &&
192               Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
193               CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
194             Prev->Type = TT_BinaryOperator;
195             LookForDecls = false;
196           }
197         }
198       }
199 
200       if (CurrentToken->Previous->is(TT_PointerOrReference) &&
201           CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
202                                                     tok::coloncolon))
203         MightBeFunctionType = true;
204       if (CurrentToken->Previous->is(TT_BinaryOperator))
205         Contexts.back().IsExpression = true;
206       if (CurrentToken->is(tok::r_paren)) {
207         if (MightBeFunctionType && CurrentToken->Next &&
208             (CurrentToken->Next->is(tok::l_paren) ||
209              (CurrentToken->Next->is(tok::l_square) &&
210               Line.MustBeDeclaration)))
211           Left->Type = TT_FunctionTypeLParen;
212         Left->MatchingParen = CurrentToken;
213         CurrentToken->MatchingParen = Left;
214 
215         if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) &&
216             Left->Previous && Left->Previous->is(tok::l_paren)) {
217           // Detect the case where macros are used to generate lambdas or
218           // function bodies, e.g.:
219           //   auto my_lambda = MARCO((Type *type, int i) { .. body .. });
220           for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) {
221             if (Tok->is(TT_BinaryOperator) &&
222                 Tok->isOneOf(tok::star, tok::amp, tok::ampamp))
223               Tok->Type = TT_PointerOrReference;
224           }
225         }
226 
227         if (StartsObjCMethodExpr) {
228           CurrentToken->Type = TT_ObjCMethodExpr;
229           if (Contexts.back().FirstObjCSelectorName) {
230             Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
231                 Contexts.back().LongestObjCSelectorName;
232           }
233         }
234 
235         if (Left->is(TT_AttributeParen))
236           CurrentToken->Type = TT_AttributeParen;
237         if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
238           CurrentToken->Type = TT_JavaAnnotation;
239         if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
240           CurrentToken->Type = TT_LeadingJavaAnnotation;
241 
242         if (!HasMultipleLines)
243           Left->PackingKind = PPK_Inconclusive;
244         else if (HasMultipleParametersOnALine)
245           Left->PackingKind = PPK_BinPacked;
246         else
247           Left->PackingKind = PPK_OnePerLine;
248 
249         next();
250         return true;
251       }
252       if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
253         return false;
254 
255       if (CurrentToken->is(tok::l_brace))
256         Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
257       if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
258           !CurrentToken->Next->HasUnescapedNewline &&
259           !CurrentToken->Next->isTrailingComment())
260         HasMultipleParametersOnALine = true;
261       if (CurrentToken->isOneOf(tok::kw_const, tok::kw_auto) ||
262           CurrentToken->isSimpleTypeSpecifier())
263         Contexts.back().IsExpression = false;
264       if (CurrentToken->isOneOf(tok::semi, tok::colon))
265         MightBeObjCForRangeLoop = false;
266       if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
267         CurrentToken->Type = TT_ObjCForIn;
268       // When we discover a 'new', we set CanBeExpression to 'false' in order to
269       // parse the type correctly. Reset that after a comma.
270       if (CurrentToken->is(tok::comma))
271         Contexts.back().CanBeExpression = true;
272 
273       FormatToken *Tok = CurrentToken;
274       if (!consumeToken())
275         return false;
276       updateParameterCount(Left, Tok);
277       if (CurrentToken && CurrentToken->HasUnescapedNewline)
278         HasMultipleLines = true;
279     }
280     return false;
281   }
282 
283   bool parseSquare() {
284     if (!CurrentToken)
285       return false;
286 
287     // A '[' could be an index subscript (after an identifier or after
288     // ')' or ']'), it could be the start of an Objective-C method
289     // expression, or it could the start of an Objective-C array literal.
290     FormatToken *Left = CurrentToken->Previous;
291     Left->ParentBracket = Contexts.back().ContextKind;
292     FormatToken *Parent = Left->getPreviousNonComment();
293     bool StartsObjCMethodExpr =
294         Style.Language == FormatStyle::LK_Cpp &&
295         Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
296         CurrentToken->isNot(tok::l_brace) &&
297         (!Parent ||
298          Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
299                          tok::kw_return, tok::kw_throw) ||
300          Parent->isUnaryOperator() ||
301          Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
302          getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
303     bool ColonFound = false;
304 
305     unsigned BindingIncrease = 1;
306     if (Left->is(TT_Unknown)) {
307       if (StartsObjCMethodExpr) {
308         Left->Type = TT_ObjCMethodExpr;
309       } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
310                  Contexts.back().ContextKind == tok::l_brace &&
311                  Parent->isOneOf(tok::l_brace, tok::comma)) {
312         Left->Type = TT_JsComputedPropertyName;
313       } else if (Style.Language == FormatStyle::LK_Proto ||
314                  (Parent &&
315                   Parent->isOneOf(TT_BinaryOperator, tok::at, tok::comma,
316                                   tok::l_paren, tok::l_square, tok::question,
317                                   tok::colon, tok::kw_return,
318                                   // Should only be relevant to JavaScript:
319                                   tok::kw_default))) {
320         Left->Type = TT_ArrayInitializerLSquare;
321       } else {
322         BindingIncrease = 10;
323         Left->Type = TT_ArraySubscriptLSquare;
324       }
325     }
326 
327     ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
328     Contexts.back().IsExpression = true;
329     Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
330 
331     while (CurrentToken) {
332       if (CurrentToken->is(tok::r_square)) {
333         if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
334             Left->is(TT_ObjCMethodExpr)) {
335           // An ObjC method call is rarely followed by an open parenthesis.
336           // FIXME: Do we incorrectly label ":" with this?
337           StartsObjCMethodExpr = false;
338           Left->Type = TT_Unknown;
339         }
340         if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
341           CurrentToken->Type = TT_ObjCMethodExpr;
342           // determineStarAmpUsage() thinks that '*' '[' is allocating an
343           // array of pointers, but if '[' starts a selector then '*' is a
344           // binary operator.
345           if (Parent && Parent->is(TT_PointerOrReference))
346             Parent->Type = TT_BinaryOperator;
347         }
348         Left->MatchingParen = CurrentToken;
349         CurrentToken->MatchingParen = Left;
350         if (Contexts.back().FirstObjCSelectorName) {
351           Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
352               Contexts.back().LongestObjCSelectorName;
353           if (Left->BlockParameterCount > 1)
354             Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
355         }
356         next();
357         return true;
358       }
359       if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
360         return false;
361       if (CurrentToken->is(tok::colon)) {
362         if (Left->is(TT_ArraySubscriptLSquare)) {
363           Left->Type = TT_ObjCMethodExpr;
364           StartsObjCMethodExpr = true;
365           Contexts.back().ColonIsObjCMethodExpr = true;
366           if (Parent && Parent->is(tok::r_paren))
367             Parent->Type = TT_CastRParen;
368         }
369         ColonFound = true;
370       }
371       if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
372           !ColonFound)
373         Left->Type = TT_ArrayInitializerLSquare;
374       FormatToken *Tok = CurrentToken;
375       if (!consumeToken())
376         return false;
377       updateParameterCount(Left, Tok);
378     }
379     return false;
380   }
381 
382   bool parseBrace() {
383     if (CurrentToken) {
384       FormatToken *Left = CurrentToken->Previous;
385       Left->ParentBracket = Contexts.back().ContextKind;
386 
387       if (Contexts.back().CaretFound)
388         Left->Type = TT_ObjCBlockLBrace;
389       Contexts.back().CaretFound = false;
390 
391       ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
392       Contexts.back().ColonIsDictLiteral = true;
393       if (Left->BlockKind == BK_BracedInit)
394         Contexts.back().IsExpression = true;
395 
396       while (CurrentToken) {
397         if (CurrentToken->is(tok::r_brace)) {
398           Left->MatchingParen = CurrentToken;
399           CurrentToken->MatchingParen = Left;
400           next();
401           return true;
402         }
403         if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
404           return false;
405         updateParameterCount(Left, CurrentToken);
406         if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) {
407           FormatToken *Previous = CurrentToken->getPreviousNonComment();
408           if (((CurrentToken->is(tok::colon) &&
409                 (!Contexts.back().ColonIsDictLiteral ||
410                  Style.Language != FormatStyle::LK_Cpp)) ||
411                Style.Language == FormatStyle::LK_Proto) &&
412               Previous->Tok.getIdentifierInfo())
413             Previous->Type = TT_SelectorName;
414           if (CurrentToken->is(tok::colon) ||
415               Style.Language == FormatStyle::LK_JavaScript)
416             Left->Type = TT_DictLiteral;
417         }
418         if (!consumeToken())
419           return false;
420       }
421     }
422     return true;
423   }
424 
425   void updateParameterCount(FormatToken *Left, FormatToken *Current) {
426     if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block)
427       ++Left->BlockParameterCount;
428     if (Current->is(tok::comma)) {
429       ++Left->ParameterCount;
430       if (!Left->Role)
431         Left->Role.reset(new CommaSeparatedList(Style));
432       Left->Role->CommaFound(Current);
433     } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
434       Left->ParameterCount = 1;
435     }
436   }
437 
438   bool parseConditional() {
439     while (CurrentToken) {
440       if (CurrentToken->is(tok::colon)) {
441         CurrentToken->Type = TT_ConditionalExpr;
442         next();
443         return true;
444       }
445       if (!consumeToken())
446         return false;
447     }
448     return false;
449   }
450 
451   bool parseTemplateDeclaration() {
452     if (CurrentToken && CurrentToken->is(tok::less)) {
453       CurrentToken->Type = TT_TemplateOpener;
454       next();
455       if (!parseAngle())
456         return false;
457       if (CurrentToken)
458         CurrentToken->Previous->ClosesTemplateDeclaration = true;
459       return true;
460     }
461     return false;
462   }
463 
464   bool consumeToken() {
465     FormatToken *Tok = CurrentToken;
466     next();
467     switch (Tok->Tok.getKind()) {
468     case tok::plus:
469     case tok::minus:
470       if (!Tok->Previous && Line.MustBeDeclaration)
471         Tok->Type = TT_ObjCMethodSpecifier;
472       break;
473     case tok::colon:
474       if (!Tok->Previous)
475         return false;
476       // Colons from ?: are handled in parseConditional().
477       if (Style.Language == FormatStyle::LK_JavaScript) {
478         if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
479             (Contexts.size() == 1 &&               // switch/case labels
480              !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
481             Contexts.back().ContextKind == tok::l_paren ||  // function params
482             Contexts.back().ContextKind == tok::l_square || // array type
483             (Contexts.size() == 1 &&
484              Line.MustBeDeclaration)) { // method/property declaration
485           Tok->Type = TT_JsTypeColon;
486           break;
487         }
488       }
489       if (Contexts.back().ColonIsDictLiteral ||
490           Style.Language == FormatStyle::LK_Proto) {
491         Tok->Type = TT_DictLiteral;
492       } else if (Contexts.back().ColonIsObjCMethodExpr ||
493                  Line.startsWith(TT_ObjCMethodSpecifier)) {
494         Tok->Type = TT_ObjCMethodExpr;
495         Tok->Previous->Type = TT_SelectorName;
496         if (Tok->Previous->ColumnWidth >
497             Contexts.back().LongestObjCSelectorName)
498           Contexts.back().LongestObjCSelectorName = Tok->Previous->ColumnWidth;
499         if (!Contexts.back().FirstObjCSelectorName)
500           Contexts.back().FirstObjCSelectorName = Tok->Previous;
501       } else if (Contexts.back().ColonIsForRangeExpr) {
502         Tok->Type = TT_RangeBasedForLoopColon;
503       } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
504         Tok->Type = TT_BitFieldColon;
505       } else if (Contexts.size() == 1 &&
506                  !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
507         if (Tok->Previous->is(tok::r_paren))
508           Tok->Type = TT_CtorInitializerColon;
509         else
510           Tok->Type = TT_InheritanceColon;
511       } else if (Tok->Previous->is(tok::identifier) && Tok->Next &&
512                  Tok->Next->isOneOf(tok::r_paren, tok::comma)) {
513         // This handles a special macro in ObjC code where selectors including
514         // the colon are passed as macro arguments.
515         Tok->Type = TT_ObjCMethodExpr;
516       } else if (Contexts.back().ContextKind == tok::l_paren) {
517         Tok->Type = TT_InlineASMColon;
518       }
519       break;
520     case tok::kw_if:
521     case tok::kw_while:
522       if (CurrentToken && CurrentToken->is(tok::l_paren)) {
523         next();
524         if (!parseParens(/*LookForDecls=*/true))
525           return false;
526       }
527       break;
528     case tok::kw_for:
529       Contexts.back().ColonIsForRangeExpr = true;
530       next();
531       if (!parseParens())
532         return false;
533       break;
534     case tok::l_paren:
535       // When faced with 'operator()()', the kw_operator handler incorrectly
536       // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
537       // the first two parens OverloadedOperators and the second l_paren an
538       // OverloadedOperatorLParen.
539       if (Tok->Previous &&
540           Tok->Previous->is(tok::r_paren) &&
541           Tok->Previous->MatchingParen &&
542           Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
543         Tok->Previous->Type = TT_OverloadedOperator;
544         Tok->Previous->MatchingParen->Type = TT_OverloadedOperator;
545         Tok->Type = TT_OverloadedOperatorLParen;
546       }
547 
548       if (!parseParens())
549         return false;
550       if (Line.MustBeDeclaration && Contexts.size() == 1 &&
551           !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
552           (!Tok->Previous ||
553            !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute,
554                                    TT_LeadingJavaAnnotation)))
555         Line.MightBeFunctionDecl = true;
556       break;
557     case tok::l_square:
558       if (!parseSquare())
559         return false;
560       break;
561     case tok::l_brace:
562       if (!parseBrace())
563         return false;
564       break;
565     case tok::less:
566       if (parseAngle()) {
567         Tok->Type = TT_TemplateOpener;
568       } else {
569         Tok->Type = TT_BinaryOperator;
570         NonTemplateLess.insert(Tok);
571         CurrentToken = Tok;
572         next();
573       }
574       break;
575     case tok::r_paren:
576     case tok::r_square:
577       return false;
578     case tok::r_brace:
579       // Lines can start with '}'.
580       if (Tok->Previous)
581         return false;
582       break;
583     case tok::greater:
584       Tok->Type = TT_BinaryOperator;
585       break;
586     case tok::kw_operator:
587       while (CurrentToken &&
588              !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
589         if (CurrentToken->isOneOf(tok::star, tok::amp))
590           CurrentToken->Type = TT_PointerOrReference;
591         consumeToken();
592         if (CurrentToken &&
593             CurrentToken->Previous->isOneOf(TT_BinaryOperator, tok::comma))
594           CurrentToken->Previous->Type = TT_OverloadedOperator;
595       }
596       if (CurrentToken) {
597         CurrentToken->Type = TT_OverloadedOperatorLParen;
598         if (CurrentToken->Previous->is(TT_BinaryOperator))
599           CurrentToken->Previous->Type = TT_OverloadedOperator;
600       }
601       break;
602     case tok::question:
603       if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
604           Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
605                              tok::r_brace)) {
606         // Question marks before semicolons, colons, etc. indicate optional
607         // types (fields, parameters), e.g.
608         //   function(x?: string, y?) {...}
609         //   class X { y?; }
610         Tok->Type = TT_JsTypeOptionalQuestion;
611         break;
612       }
613       // Declarations cannot be conditional expressions, this can only be part
614       // of a type declaration.
615       if (Line.MustBeDeclaration &&
616           Style.Language == FormatStyle::LK_JavaScript)
617         break;
618       parseConditional();
619       break;
620     case tok::kw_template:
621       parseTemplateDeclaration();
622       break;
623     case tok::comma:
624       if (Contexts.back().InCtorInitializer)
625         Tok->Type = TT_CtorInitializerComma;
626       else if (Contexts.back().FirstStartOfName &&
627                (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
628         Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
629         Line.IsMultiVariableDeclStmt = true;
630       }
631       if (Contexts.back().IsForEachMacro)
632         Contexts.back().IsExpression = true;
633       break;
634     default:
635       break;
636     }
637     return true;
638   }
639 
640   void parseIncludeDirective() {
641     if (CurrentToken && CurrentToken->is(tok::less)) {
642       next();
643       while (CurrentToken) {
644         if (CurrentToken->isNot(tok::comment) || CurrentToken->Next)
645           CurrentToken->Type = TT_ImplicitStringLiteral;
646         next();
647       }
648     }
649   }
650 
651   void parseWarningOrError() {
652     next();
653     // We still want to format the whitespace left of the first token of the
654     // warning or error.
655     next();
656     while (CurrentToken) {
657       CurrentToken->Type = TT_ImplicitStringLiteral;
658       next();
659     }
660   }
661 
662   void parsePragma() {
663     next(); // Consume "pragma".
664     if (CurrentToken &&
665         CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
666       bool IsMark = CurrentToken->is(Keywords.kw_mark);
667       next(); // Consume "mark".
668       next(); // Consume first token (so we fix leading whitespace).
669       while (CurrentToken) {
670         if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
671           CurrentToken->Type = TT_ImplicitStringLiteral;
672         next();
673       }
674     }
675   }
676 
677   LineType parsePreprocessorDirective() {
678     LineType Type = LT_PreprocessorDirective;
679     next();
680     if (!CurrentToken)
681       return Type;
682     if (CurrentToken->Tok.is(tok::numeric_constant)) {
683       CurrentToken->SpacesRequiredBefore = 1;
684       return Type;
685     }
686     // Hashes in the middle of a line can lead to any strange token
687     // sequence.
688     if (!CurrentToken->Tok.getIdentifierInfo())
689       return Type;
690     switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
691     case tok::pp_include:
692     case tok::pp_include_next:
693     case tok::pp_import:
694       next();
695       parseIncludeDirective();
696       Type = LT_ImportStatement;
697       break;
698     case tok::pp_error:
699     case tok::pp_warning:
700       parseWarningOrError();
701       break;
702     case tok::pp_pragma:
703       parsePragma();
704       break;
705     case tok::pp_if:
706     case tok::pp_elif:
707       Contexts.back().IsExpression = true;
708       parseLine();
709       break;
710     default:
711       break;
712     }
713     while (CurrentToken)
714       next();
715     return Type;
716   }
717 
718 public:
719   LineType parseLine() {
720     NonTemplateLess.clear();
721     if (CurrentToken->is(tok::hash))
722       return parsePreprocessorDirective();
723 
724     // Directly allow to 'import <string-literal>' to support protocol buffer
725     // definitions (code.google.com/p/protobuf) or missing "#" (either way we
726     // should not break the line).
727     IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
728     if ((Style.Language == FormatStyle::LK_Java &&
729          CurrentToken->is(Keywords.kw_package)) ||
730         (Info && Info->getPPKeywordID() == tok::pp_import &&
731          CurrentToken->Next &&
732          CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
733                                      tok::kw_static))) {
734       next();
735       parseIncludeDirective();
736       return LT_ImportStatement;
737     }
738 
739     // If this line starts and ends in '<' and '>', respectively, it is likely
740     // part of "#define <a/b.h>".
741     if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
742       parseIncludeDirective();
743       return LT_ImportStatement;
744     }
745 
746     // In .proto files, top-level options are very similar to import statements
747     // and should not be line-wrapped.
748     if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
749         CurrentToken->is(Keywords.kw_option)) {
750       next();
751       if (CurrentToken && CurrentToken->is(tok::identifier))
752         return LT_ImportStatement;
753     }
754 
755     bool KeywordVirtualFound = false;
756     bool ImportStatement = false;
757     while (CurrentToken) {
758       if (CurrentToken->is(tok::kw_virtual))
759         KeywordVirtualFound = true;
760       if (isImportStatement(*CurrentToken))
761         ImportStatement = true;
762       if (!consumeToken())
763         return LT_Invalid;
764     }
765     if (KeywordVirtualFound)
766       return LT_VirtualFunctionDecl;
767     if (ImportStatement)
768       return LT_ImportStatement;
769 
770     if (Line.startsWith(TT_ObjCMethodSpecifier)) {
771       if (Contexts.back().FirstObjCSelectorName)
772         Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
773             Contexts.back().LongestObjCSelectorName;
774       return LT_ObjCMethodDecl;
775     }
776 
777     return LT_Other;
778   }
779 
780 private:
781   bool isImportStatement(const FormatToken &Tok) {
782     // FIXME: Closure-library specific stuff should not be hard-coded but be
783     // configurable.
784     return Style.Language == FormatStyle::LK_JavaScript &&
785            Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
786            Tok.Next->Next && (Tok.Next->Next->TokenText == "module" ||
787                               Tok.Next->Next->TokenText == "provide" ||
788                               Tok.Next->Next->TokenText == "require" ||
789                               Tok.Next->Next->TokenText == "setTestOnly") &&
790            Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
791   }
792 
793   void resetTokenMetadata(FormatToken *Token) {
794     if (!Token)
795       return;
796 
797     // Reset token type in case we have already looked at it and then
798     // recovered from an error (e.g. failure to find the matching >).
799     if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
800                                TT_FunctionLBrace, TT_ImplicitStringLiteral,
801                                TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
802                                TT_RegexLiteral))
803       CurrentToken->Type = TT_Unknown;
804     CurrentToken->Role.reset();
805     CurrentToken->MatchingParen = nullptr;
806     CurrentToken->FakeLParens.clear();
807     CurrentToken->FakeRParens = 0;
808   }
809 
810   void next() {
811     if (CurrentToken) {
812       CurrentToken->NestingLevel = Contexts.size() - 1;
813       CurrentToken->BindingStrength = Contexts.back().BindingStrength;
814       modifyContext(*CurrentToken);
815       determineTokenType(*CurrentToken);
816       CurrentToken = CurrentToken->Next;
817     }
818 
819     resetTokenMetadata(CurrentToken);
820   }
821 
822   /// \brief A struct to hold information valid in a specific context, e.g.
823   /// a pair of parenthesis.
824   struct Context {
825     Context(tok::TokenKind ContextKind, unsigned BindingStrength,
826             bool IsExpression)
827         : ContextKind(ContextKind), BindingStrength(BindingStrength),
828           IsExpression(IsExpression) {}
829 
830     tok::TokenKind ContextKind;
831     unsigned BindingStrength;
832     bool IsExpression;
833     unsigned LongestObjCSelectorName = 0;
834     bool ColonIsForRangeExpr = false;
835     bool ColonIsDictLiteral = false;
836     bool ColonIsObjCMethodExpr = false;
837     FormatToken *FirstObjCSelectorName = nullptr;
838     FormatToken *FirstStartOfName = nullptr;
839     bool CanBeExpression = true;
840     bool InTemplateArgument = false;
841     bool InCtorInitializer = false;
842     bool CaretFound = false;
843     bool IsForEachMacro = false;
844   };
845 
846   /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
847   /// of each instance.
848   struct ScopedContextCreator {
849     AnnotatingParser &P;
850 
851     ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
852                          unsigned Increase)
853         : P(P) {
854       P.Contexts.push_back(Context(ContextKind,
855                                    P.Contexts.back().BindingStrength + Increase,
856                                    P.Contexts.back().IsExpression));
857     }
858 
859     ~ScopedContextCreator() { P.Contexts.pop_back(); }
860   };
861 
862   void modifyContext(const FormatToken &Current) {
863     if (Current.getPrecedence() == prec::Assignment &&
864         !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
865         (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
866       Contexts.back().IsExpression = true;
867       if (!Line.startsWith(TT_UnaryOperator)) {
868         for (FormatToken *Previous = Current.Previous;
869              Previous && Previous->Previous &&
870              !Previous->Previous->isOneOf(tok::comma, tok::semi);
871              Previous = Previous->Previous) {
872           if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
873             Previous = Previous->MatchingParen;
874             if (!Previous)
875               break;
876           }
877           if (Previous->opensScope())
878             break;
879           if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
880               Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
881               Previous->Previous && Previous->Previous->isNot(tok::equal))
882             Previous->Type = TT_PointerOrReference;
883         }
884       }
885     } else if (Current.is(tok::lessless) &&
886                (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
887       Contexts.back().IsExpression = true;
888     } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
889       Contexts.back().IsExpression = true;
890     } else if (Current.is(TT_TrailingReturnArrow)) {
891       Contexts.back().IsExpression = false;
892     } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
893       Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
894     } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
895       for (FormatToken *Previous = Current.Previous;
896            Previous && Previous->isOneOf(tok::star, tok::amp);
897            Previous = Previous->Previous)
898         Previous->Type = TT_PointerOrReference;
899       if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
900         Contexts.back().IsExpression = false;
901     } else if (Current.Previous &&
902                Current.Previous->is(TT_CtorInitializerColon)) {
903       Contexts.back().IsExpression = true;
904       Contexts.back().InCtorInitializer = true;
905     } else if (Current.is(tok::kw_new)) {
906       Contexts.back().CanBeExpression = false;
907     } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
908       // This should be the condition or increment in a for-loop.
909       Contexts.back().IsExpression = true;
910     }
911   }
912 
913   void determineTokenType(FormatToken &Current) {
914     if (!Current.is(TT_Unknown))
915       // The token type is already known.
916       return;
917 
918     // Line.MightBeFunctionDecl can only be true after the parentheses of a
919     // function declaration have been found. In this case, 'Current' is a
920     // trailing token of this declaration and thus cannot be a name.
921     if (Current.is(Keywords.kw_instanceof)) {
922       Current.Type = TT_BinaryOperator;
923     } else if (isStartOfName(Current) &&
924                (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
925       Contexts.back().FirstStartOfName = &Current;
926       Current.Type = TT_StartOfName;
927     } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
928       AutoFound = true;
929     } else if (Current.is(tok::arrow) &&
930                Style.Language == FormatStyle::LK_Java) {
931       Current.Type = TT_LambdaArrow;
932     } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
933                Current.NestingLevel == 0) {
934       Current.Type = TT_TrailingReturnArrow;
935     } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
936       Current.Type =
937           determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
938                                              Contexts.back().IsExpression,
939                                 Contexts.back().InTemplateArgument);
940     } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
941       Current.Type = determinePlusMinusCaretUsage(Current);
942       if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
943         Contexts.back().CaretFound = true;
944     } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
945       Current.Type = determineIncrementUsage(Current);
946     } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
947       Current.Type = TT_UnaryOperator;
948     } else if (Current.is(tok::question)) {
949       if (Style.Language == FormatStyle::LK_JavaScript &&
950           Line.MustBeDeclaration) {
951         // In JavaScript, `interface X { foo?(): bar; }` is an optional method
952         // on the interface, not a ternary expression.
953         Current.Type = TT_JsTypeOptionalQuestion;
954       } else {
955         Current.Type = TT_ConditionalExpr;
956       }
957     } else if (Current.isBinaryOperator() &&
958                (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
959       Current.Type = TT_BinaryOperator;
960     } else if (Current.is(tok::comment)) {
961       if (Current.TokenText.startswith("/*")) {
962         if (Current.TokenText.endswith("*/"))
963           Current.Type = TT_BlockComment;
964         else
965           // The lexer has for some reason determined a comment here. But we
966           // cannot really handle it, if it isn't properly terminated.
967           Current.Tok.setKind(tok::unknown);
968       } else {
969         Current.Type = TT_LineComment;
970       }
971     } else if (Current.is(tok::r_paren)) {
972       if (rParenEndsCast(Current))
973         Current.Type = TT_CastRParen;
974       if (Current.MatchingParen && Current.Next &&
975           !Current.Next->isBinaryOperator() &&
976           !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace))
977         if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
978           if (BeforeParen->is(tok::identifier) &&
979               BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
980               (!BeforeParen->Previous ||
981                BeforeParen->Previous->ClosesTemplateDeclaration))
982             Current.Type = TT_FunctionAnnotationRParen;
983     } else if (Current.is(tok::at) && Current.Next) {
984       if (Current.Next->isStringLiteral()) {
985         Current.Type = TT_ObjCStringLiteral;
986       } else {
987         switch (Current.Next->Tok.getObjCKeywordID()) {
988         case tok::objc_interface:
989         case tok::objc_implementation:
990         case tok::objc_protocol:
991           Current.Type = TT_ObjCDecl;
992           break;
993         case tok::objc_property:
994           Current.Type = TT_ObjCProperty;
995           break;
996         default:
997           break;
998         }
999       }
1000     } else if (Current.is(tok::period)) {
1001       FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1002       if (PreviousNoComment &&
1003           PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1004         Current.Type = TT_DesignatedInitializerPeriod;
1005       else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1006                Current.Previous->isOneOf(TT_JavaAnnotation,
1007                                          TT_LeadingJavaAnnotation)) {
1008         Current.Type = Current.Previous->Type;
1009       }
1010     } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
1011                Current.Previous &&
1012                !Current.Previous->isOneOf(tok::equal, tok::at) &&
1013                Line.MightBeFunctionDecl && Contexts.size() == 1) {
1014       // Line.MightBeFunctionDecl can only be true after the parentheses of a
1015       // function declaration have been found.
1016       Current.Type = TT_TrailingAnnotation;
1017     } else if ((Style.Language == FormatStyle::LK_Java ||
1018                 Style.Language == FormatStyle::LK_JavaScript) &&
1019                Current.Previous) {
1020       if (Current.Previous->is(tok::at) &&
1021           Current.isNot(Keywords.kw_interface)) {
1022         const FormatToken &AtToken = *Current.Previous;
1023         const FormatToken *Previous = AtToken.getPreviousNonComment();
1024         if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1025           Current.Type = TT_LeadingJavaAnnotation;
1026         else
1027           Current.Type = TT_JavaAnnotation;
1028       } else if (Current.Previous->is(tok::period) &&
1029                  Current.Previous->isOneOf(TT_JavaAnnotation,
1030                                            TT_LeadingJavaAnnotation)) {
1031         Current.Type = Current.Previous->Type;
1032       }
1033     }
1034   }
1035 
1036   /// \brief Take a guess at whether \p Tok starts a name of a function or
1037   /// variable declaration.
1038   ///
1039   /// This is a heuristic based on whether \p Tok is an identifier following
1040   /// something that is likely a type.
1041   bool isStartOfName(const FormatToken &Tok) {
1042     if (Tok.isNot(tok::identifier) || !Tok.Previous)
1043       return false;
1044 
1045     if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof))
1046       return false;
1047     if (Style.Language == FormatStyle::LK_JavaScript &&
1048         Tok.Previous->is(Keywords.kw_in))
1049       return false;
1050 
1051     // Skip "const" as it does not have an influence on whether this is a name.
1052     FormatToken *PreviousNotConst = Tok.Previous;
1053     while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1054       PreviousNotConst = PreviousNotConst->Previous;
1055 
1056     if (!PreviousNotConst)
1057       return false;
1058 
1059     bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1060                        PreviousNotConst->Previous &&
1061                        PreviousNotConst->Previous->is(tok::hash);
1062 
1063     if (PreviousNotConst->is(TT_TemplateCloser))
1064       return PreviousNotConst && PreviousNotConst->MatchingParen &&
1065              PreviousNotConst->MatchingParen->Previous &&
1066              PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1067              PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1068 
1069     if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1070         PreviousNotConst->MatchingParen->Previous &&
1071         PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1072       return true;
1073 
1074     return (!IsPPKeyword &&
1075             PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1076            PreviousNotConst->is(TT_PointerOrReference) ||
1077            PreviousNotConst->isSimpleTypeSpecifier();
1078   }
1079 
1080   /// \brief Determine whether ')' is ending a cast.
1081   bool rParenEndsCast(const FormatToken &Tok) {
1082     // C-style casts are only used in C++ and Java.
1083     if (Style.Language != FormatStyle::LK_Cpp &&
1084         Style.Language != FormatStyle::LK_Java)
1085       return false;
1086 
1087     // Empty parens aren't casts and there are no casts at the end of the line.
1088     if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1089       return false;
1090 
1091     FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1092     if (LeftOfParens) {
1093       // If there is an opening parenthesis left of the current parentheses,
1094       // look past it as these might be chained casts.
1095       if (LeftOfParens->is(tok::r_paren)) {
1096         if (!LeftOfParens->MatchingParen ||
1097             !LeftOfParens->MatchingParen->Previous)
1098           return false;
1099         LeftOfParens = LeftOfParens->MatchingParen->Previous;
1100       }
1101 
1102       // If there is an identifier (or with a few exceptions a keyword) right
1103       // before the parentheses, this is unlikely to be a cast.
1104       if (LeftOfParens->Tok.getIdentifierInfo() &&
1105           !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1106                                  tok::kw_delete))
1107         return false;
1108 
1109       // Certain other tokens right before the parentheses are also signals that
1110       // this cannot be a cast.
1111       if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1112                                 TT_TemplateCloser))
1113         return false;
1114     }
1115 
1116     if (Tok.Next->is(tok::question))
1117       return false;
1118 
1119     // As Java has no function types, a "(" after the ")" likely means that this
1120     // is a cast.
1121     if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1122       return true;
1123 
1124     // If a (non-string) literal follows, this is likely a cast.
1125     if (Tok.Next->isNot(tok::string_literal) &&
1126         (Tok.Next->Tok.isLiteral() ||
1127          Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1128       return true;
1129 
1130     // Heuristically try to determine whether the parentheses contain a type.
1131     bool ParensAreType =
1132         !Tok.Previous ||
1133         Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1134         Tok.Previous->isSimpleTypeSpecifier();
1135     bool ParensCouldEndDecl =
1136         Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1137     if (ParensAreType && !ParensCouldEndDecl)
1138       return true;
1139 
1140     // At this point, we heuristically assume that there are no casts at the
1141     // start of the line. We assume that we have found most cases where there
1142     // are by the logic above, e.g. "(void)x;".
1143     if (!LeftOfParens)
1144       return false;
1145 
1146     // If the following token is an identifier, this is a cast. All cases where
1147     // this can be something else are handled above.
1148     if (Tok.Next->is(tok::identifier))
1149       return true;
1150 
1151     if (!Tok.Next->Next)
1152       return false;
1153 
1154     // If the next token after the parenthesis is a unary operator, assume
1155     // that this is cast, unless there are unexpected tokens inside the
1156     // parenthesis.
1157     bool NextIsUnary =
1158         Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1159     if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1160         !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1161       return false;
1162     // Search for unexpected tokens.
1163     for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1164          Prev = Prev->Previous) {
1165       if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1166         return false;
1167     }
1168     return true;
1169   }
1170 
1171   /// \brief Return the type of the given token assuming it is * or &.
1172   TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1173                                   bool InTemplateArgument) {
1174     if (Style.Language == FormatStyle::LK_JavaScript)
1175       return TT_BinaryOperator;
1176 
1177     const FormatToken *PrevToken = Tok.getPreviousNonComment();
1178     if (!PrevToken)
1179       return TT_UnaryOperator;
1180 
1181     const FormatToken *NextToken = Tok.getNextNonComment();
1182     if (!NextToken ||
1183         NextToken->isOneOf(tok::arrow, Keywords.kw_final,
1184                            Keywords.kw_override) ||
1185         (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1186       return TT_PointerOrReference;
1187 
1188     if (PrevToken->is(tok::coloncolon))
1189       return TT_PointerOrReference;
1190 
1191     if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1192                            tok::comma, tok::semi, tok::kw_return, tok::colon,
1193                            tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1194         PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1195                            TT_UnaryOperator, TT_CastRParen))
1196       return TT_UnaryOperator;
1197 
1198     if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1199       return TT_PointerOrReference;
1200     if (NextToken->is(tok::kw_operator) && !IsExpression)
1201       return TT_PointerOrReference;
1202     if (NextToken->isOneOf(tok::comma, tok::semi))
1203       return TT_PointerOrReference;
1204 
1205     if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1206         PrevToken->MatchingParen->Previous &&
1207         PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1208                                                     tok::kw_decltype))
1209       return TT_PointerOrReference;
1210 
1211     if (PrevToken->Tok.isLiteral() ||
1212         PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1213                            tok::kw_false, tok::r_brace) ||
1214         NextToken->Tok.isLiteral() ||
1215         NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1216         NextToken->isUnaryOperator() ||
1217         // If we know we're in a template argument, there are no named
1218         // declarations. Thus, having an identifier on the right-hand side
1219         // indicates a binary operator.
1220         (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1221       return TT_BinaryOperator;
1222 
1223     // "&&(" is quite unlikely to be two successive unary "&".
1224     if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1225       return TT_BinaryOperator;
1226 
1227     // This catches some cases where evaluation order is used as control flow:
1228     //   aaa && aaa->f();
1229     const FormatToken *NextNextToken = NextToken->getNextNonComment();
1230     if (NextNextToken && NextNextToken->is(tok::arrow))
1231       return TT_BinaryOperator;
1232 
1233     // It is very unlikely that we are going to find a pointer or reference type
1234     // definition on the RHS of an assignment.
1235     if (IsExpression && !Contexts.back().CaretFound)
1236       return TT_BinaryOperator;
1237 
1238     return TT_PointerOrReference;
1239   }
1240 
1241   TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1242     const FormatToken *PrevToken = Tok.getPreviousNonComment();
1243     if (!PrevToken || PrevToken->is(TT_CastRParen))
1244       return TT_UnaryOperator;
1245 
1246     // Use heuristics to recognize unary operators.
1247     if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1248                            tok::question, tok::colon, tok::kw_return,
1249                            tok::kw_case, tok::at, tok::l_brace))
1250       return TT_UnaryOperator;
1251 
1252     // There can't be two consecutive binary operators.
1253     if (PrevToken->is(TT_BinaryOperator))
1254       return TT_UnaryOperator;
1255 
1256     // Fall back to marking the token as binary operator.
1257     return TT_BinaryOperator;
1258   }
1259 
1260   /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1261   TokenType determineIncrementUsage(const FormatToken &Tok) {
1262     const FormatToken *PrevToken = Tok.getPreviousNonComment();
1263     if (!PrevToken || PrevToken->is(TT_CastRParen))
1264       return TT_UnaryOperator;
1265     if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1266       return TT_TrailingUnaryOperator;
1267 
1268     return TT_UnaryOperator;
1269   }
1270 
1271   SmallVector<Context, 8> Contexts;
1272 
1273   const FormatStyle &Style;
1274   AnnotatedLine &Line;
1275   FormatToken *CurrentToken;
1276   bool AutoFound;
1277   const AdditionalKeywords &Keywords;
1278 
1279   // Set of "<" tokens that do not open a template parameter list. If parseAngle
1280   // determines that a specific token can't be a template opener, it will make
1281   // same decision irrespective of the decisions for tokens leading up to it.
1282   // Store this information to prevent this from causing exponential runtime.
1283   llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1284 };
1285 
1286 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1287 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1288 
1289 /// \brief Parses binary expressions by inserting fake parenthesis based on
1290 /// operator precedence.
1291 class ExpressionParser {
1292 public:
1293   ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1294                    AnnotatedLine &Line)
1295       : Style(Style), Keywords(Keywords), Current(Line.First) {}
1296 
1297   /// \brief Parse expressions with the given operatore precedence.
1298   void parse(int Precedence = 0) {
1299     // Skip 'return' and ObjC selector colons as they are not part of a binary
1300     // expression.
1301     while (Current && (Current->is(tok::kw_return) ||
1302                        (Current->is(tok::colon) &&
1303                         Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1304       next();
1305 
1306     if (!Current || Precedence > PrecedenceArrowAndPeriod)
1307       return;
1308 
1309     // Conditional expressions need to be parsed separately for proper nesting.
1310     if (Precedence == prec::Conditional) {
1311       parseConditionalExpr();
1312       return;
1313     }
1314 
1315     // Parse unary operators, which all have a higher precedence than binary
1316     // operators.
1317     if (Precedence == PrecedenceUnaryOperator) {
1318       parseUnaryOperator();
1319       return;
1320     }
1321 
1322     FormatToken *Start = Current;
1323     FormatToken *LatestOperator = nullptr;
1324     unsigned OperatorIndex = 0;
1325 
1326     while (Current) {
1327       // Consume operators with higher precedence.
1328       parse(Precedence + 1);
1329 
1330       int CurrentPrecedence = getCurrentPrecedence();
1331 
1332       if (Current && Current->is(TT_SelectorName) &&
1333           Precedence == CurrentPrecedence) {
1334         if (LatestOperator)
1335           addFakeParenthesis(Start, prec::Level(Precedence));
1336         Start = Current;
1337       }
1338 
1339       // At the end of the line or when an operator with higher precedence is
1340       // found, insert fake parenthesis and return.
1341       if (!Current || (Current->closesScope() && Current->MatchingParen) ||
1342           (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1343           (CurrentPrecedence == prec::Conditional &&
1344            Precedence == prec::Assignment && Current->is(tok::colon))) {
1345         break;
1346       }
1347 
1348       // Consume scopes: (), [], <> and {}
1349       if (Current->opensScope()) {
1350         while (Current && !Current->closesScope()) {
1351           next();
1352           parse();
1353         }
1354         next();
1355       } else {
1356         // Operator found.
1357         if (CurrentPrecedence == Precedence) {
1358           if (LatestOperator)
1359             LatestOperator->NextOperator = Current;
1360           LatestOperator = Current;
1361           Current->OperatorIndex = OperatorIndex;
1362           ++OperatorIndex;
1363         }
1364         next(/*SkipPastLeadingComments=*/Precedence > 0);
1365       }
1366     }
1367 
1368     if (LatestOperator && (Current || Precedence > 0)) {
1369       // LatestOperator->LastOperator = true;
1370       if (Precedence == PrecedenceArrowAndPeriod) {
1371         // Call expressions don't have a binary operator precedence.
1372         addFakeParenthesis(Start, prec::Unknown);
1373       } else {
1374         addFakeParenthesis(Start, prec::Level(Precedence));
1375       }
1376     }
1377   }
1378 
1379 private:
1380   /// \brief Gets the precedence (+1) of the given token for binary operators
1381   /// and other tokens that we treat like binary operators.
1382   int getCurrentPrecedence() {
1383     if (Current) {
1384       const FormatToken *NextNonComment = Current->getNextNonComment();
1385       if (Current->is(TT_ConditionalExpr))
1386         return prec::Conditional;
1387       if (NextNonComment && NextNonComment->is(tok::colon) &&
1388           NextNonComment->is(TT_DictLiteral))
1389         return prec::Comma;
1390       if (Current->is(TT_LambdaArrow))
1391         return prec::Comma;
1392       if (Current->is(TT_JsFatArrow))
1393         return prec::Assignment;
1394       if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName,
1395                            TT_JsComputedPropertyName) ||
1396           (Current->is(tok::comment) && NextNonComment &&
1397            NextNonComment->is(TT_SelectorName)))
1398         return 0;
1399       if (Current->is(TT_RangeBasedForLoopColon))
1400         return prec::Comma;
1401       if ((Style.Language == FormatStyle::LK_Java ||
1402            Style.Language == FormatStyle::LK_JavaScript) &&
1403           Current->is(Keywords.kw_instanceof))
1404         return prec::Relational;
1405       if (Style.Language == FormatStyle::LK_JavaScript &&
1406           Current->is(Keywords.kw_in))
1407         return prec::Relational;
1408       if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1409         return Current->getPrecedence();
1410       if (Current->isOneOf(tok::period, tok::arrow))
1411         return PrecedenceArrowAndPeriod;
1412       if (Style.Language == FormatStyle::LK_Java &&
1413           Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1414                            Keywords.kw_throws))
1415         return 0;
1416     }
1417     return -1;
1418   }
1419 
1420   void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1421     Start->FakeLParens.push_back(Precedence);
1422     if (Precedence > prec::Unknown)
1423       Start->StartsBinaryExpression = true;
1424     if (Current) {
1425       FormatToken *Previous = Current->Previous;
1426       while (Previous->is(tok::comment) && Previous->Previous)
1427         Previous = Previous->Previous;
1428       ++Previous->FakeRParens;
1429       if (Precedence > prec::Unknown)
1430         Previous->EndsBinaryExpression = true;
1431     }
1432   }
1433 
1434   /// \brief Parse unary operator expressions and surround them with fake
1435   /// parentheses if appropriate.
1436   void parseUnaryOperator() {
1437     if (!Current || Current->isNot(TT_UnaryOperator)) {
1438       parse(PrecedenceArrowAndPeriod);
1439       return;
1440     }
1441 
1442     FormatToken *Start = Current;
1443     next();
1444     parseUnaryOperator();
1445 
1446     // The actual precedence doesn't matter.
1447     addFakeParenthesis(Start, prec::Unknown);
1448   }
1449 
1450   void parseConditionalExpr() {
1451     while (Current && Current->isTrailingComment()) {
1452       next();
1453     }
1454     FormatToken *Start = Current;
1455     parse(prec::LogicalOr);
1456     if (!Current || !Current->is(tok::question))
1457       return;
1458     next();
1459     parse(prec::Assignment);
1460     if (!Current || Current->isNot(TT_ConditionalExpr))
1461       return;
1462     next();
1463     parse(prec::Assignment);
1464     addFakeParenthesis(Start, prec::Conditional);
1465   }
1466 
1467   void next(bool SkipPastLeadingComments = true) {
1468     if (Current)
1469       Current = Current->Next;
1470     while (Current &&
1471            (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1472            Current->isTrailingComment())
1473       Current = Current->Next;
1474   }
1475 
1476   const FormatStyle &Style;
1477   const AdditionalKeywords &Keywords;
1478   FormatToken *Current;
1479 };
1480 
1481 } // end anonymous namespace
1482 
1483 void TokenAnnotator::setCommentLineLevels(
1484     SmallVectorImpl<AnnotatedLine *> &Lines) {
1485   const AnnotatedLine *NextNonCommentLine = nullptr;
1486   for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1487                                                           E = Lines.rend();
1488        I != E; ++I) {
1489     if (NextNonCommentLine && (*I)->First->is(tok::comment) &&
1490         (*I)->First->Next == nullptr)
1491       (*I)->Level = NextNonCommentLine->Level;
1492     else
1493       NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1494 
1495     setCommentLineLevels((*I)->Children);
1496   }
1497 }
1498 
1499 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1500   for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1501                                                   E = Line.Children.end();
1502        I != E; ++I) {
1503     annotate(**I);
1504   }
1505   AnnotatingParser Parser(Style, Line, Keywords);
1506   Line.Type = Parser.parseLine();
1507   if (Line.Type == LT_Invalid)
1508     return;
1509 
1510   ExpressionParser ExprParser(Style, Keywords, Line);
1511   ExprParser.parse();
1512 
1513   if (Line.startsWith(TT_ObjCMethodSpecifier))
1514     Line.Type = LT_ObjCMethodDecl;
1515   else if (Line.startsWith(TT_ObjCDecl))
1516     Line.Type = LT_ObjCDecl;
1517   else if (Line.startsWith(TT_ObjCProperty))
1518     Line.Type = LT_ObjCProperty;
1519 
1520   Line.First->SpacesRequiredBefore = 1;
1521   Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1522 }
1523 
1524 // This function heuristically determines whether 'Current' starts the name of a
1525 // function declaration.
1526 static bool isFunctionDeclarationName(const FormatToken &Current) {
1527   auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1528     for (; Next; Next = Next->Next) {
1529       if (Next->is(TT_OverloadedOperatorLParen))
1530         return Next;
1531       if (Next->is(TT_OverloadedOperator))
1532         continue;
1533       if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1534         // For 'new[]' and 'delete[]'.
1535         if (Next->Next && Next->Next->is(tok::l_square) &&
1536             Next->Next->Next && Next->Next->Next->is(tok::r_square))
1537           Next = Next->Next->Next;
1538         continue;
1539       }
1540 
1541       break;
1542     }
1543     return nullptr;
1544   };
1545 
1546   const FormatToken *Next = Current.Next;
1547   if (Current.is(tok::kw_operator)) {
1548     if (Current.Previous && Current.Previous->is(tok::coloncolon))
1549       return false;
1550     Next = skipOperatorName(Next);
1551   } else {
1552     if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1553       return false;
1554     for (; Next; Next = Next->Next) {
1555       if (Next->is(TT_TemplateOpener)) {
1556         Next = Next->MatchingParen;
1557       } else if (Next->is(tok::coloncolon)) {
1558         Next = Next->Next;
1559         if (!Next)
1560           return false;
1561         if (Next->is(tok::kw_operator)) {
1562           Next = skipOperatorName(Next->Next);
1563           break;
1564         }
1565         if (!Next->is(tok::identifier))
1566           return false;
1567       } else if (Next->is(tok::l_paren)) {
1568         break;
1569       } else {
1570         return false;
1571       }
1572     }
1573   }
1574 
1575   if (!Next || !Next->is(tok::l_paren))
1576     return false;
1577   if (Next->Next == Next->MatchingParen)
1578     return true;
1579   for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1580        Tok = Tok->Next) {
1581     if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1582         Tok->isOneOf(TT_PointerOrReference, TT_StartOfName))
1583       return true;
1584     if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1585         Tok->Tok.isLiteral())
1586       return false;
1587   }
1588   return false;
1589 }
1590 
1591 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1592   assert(Line.MightBeFunctionDecl);
1593 
1594   if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1595        Style.AlwaysBreakAfterReturnType ==
1596            FormatStyle::RTBS_TopLevelDefinitions) &&
1597       Line.Level > 0)
1598     return false;
1599 
1600   switch (Style.AlwaysBreakAfterReturnType) {
1601   case FormatStyle::RTBS_None:
1602     return false;
1603   case FormatStyle::RTBS_All:
1604   case FormatStyle::RTBS_TopLevel:
1605     return true;
1606   case FormatStyle::RTBS_AllDefinitions:
1607   case FormatStyle::RTBS_TopLevelDefinitions:
1608     return Line.mightBeFunctionDefinition();
1609   }
1610 
1611   return false;
1612 }
1613 
1614 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1615   for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1616                                                   E = Line.Children.end();
1617        I != E; ++I) {
1618     calculateFormattingInformation(**I);
1619   }
1620 
1621   Line.First->TotalLength =
1622       Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1623   if (!Line.First->Next)
1624     return;
1625   FormatToken *Current = Line.First->Next;
1626   bool InFunctionDecl = Line.MightBeFunctionDecl;
1627   while (Current) {
1628     if (isFunctionDeclarationName(*Current))
1629       Current->Type = TT_FunctionDeclarationName;
1630     if (Current->is(TT_LineComment)) {
1631       if (Current->Previous->BlockKind == BK_BracedInit &&
1632           Current->Previous->opensScope())
1633         Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1634       else
1635         Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1636 
1637       // If we find a trailing comment, iterate backwards to determine whether
1638       // it seems to relate to a specific parameter. If so, break before that
1639       // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1640       // to the previous line in:
1641       //   SomeFunction(a,
1642       //                b, // comment
1643       //                c);
1644       if (!Current->HasUnescapedNewline) {
1645         for (FormatToken *Parameter = Current->Previous; Parameter;
1646              Parameter = Parameter->Previous) {
1647           if (Parameter->isOneOf(tok::comment, tok::r_brace))
1648             break;
1649           if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1650             if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1651                 Parameter->HasUnescapedNewline)
1652               Parameter->MustBreakBefore = true;
1653             break;
1654           }
1655         }
1656       }
1657     } else if (Current->SpacesRequiredBefore == 0 &&
1658                spaceRequiredBefore(Line, *Current)) {
1659       Current->SpacesRequiredBefore = 1;
1660     }
1661 
1662     Current->MustBreakBefore =
1663         Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1664 
1665     if (!Current->MustBreakBefore && InFunctionDecl &&
1666         Current->is(TT_FunctionDeclarationName))
1667       Current->MustBreakBefore = mustBreakForReturnType(Line);
1668 
1669     Current->CanBreakBefore =
1670         Current->MustBreakBefore || canBreakBefore(Line, *Current);
1671     unsigned ChildSize = 0;
1672     if (Current->Previous->Children.size() == 1) {
1673       FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1674       ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1675                                                   : LastOfChild.TotalLength + 1;
1676     }
1677     const FormatToken *Prev = Current->Previous;
1678     if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1679         (Prev->Children.size() == 1 &&
1680          Prev->Children[0]->First->MustBreakBefore) ||
1681         Current->IsMultiline)
1682       Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1683     else
1684       Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1685                              ChildSize + Current->SpacesRequiredBefore;
1686 
1687     if (Current->is(TT_CtorInitializerColon))
1688       InFunctionDecl = false;
1689 
1690     // FIXME: Only calculate this if CanBreakBefore is true once static
1691     // initializers etc. are sorted out.
1692     // FIXME: Move magic numbers to a better place.
1693     Current->SplitPenalty = 20 * Current->BindingStrength +
1694                             splitPenalty(Line, *Current, InFunctionDecl);
1695 
1696     Current = Current->Next;
1697   }
1698 
1699   calculateUnbreakableTailLengths(Line);
1700   for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1701     if (Current->Role)
1702       Current->Role->precomputeFormattingInfos(Current);
1703   }
1704 
1705   DEBUG({ printDebugInfo(Line); });
1706 }
1707 
1708 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1709   unsigned UnbreakableTailLength = 0;
1710   FormatToken *Current = Line.Last;
1711   while (Current) {
1712     Current->UnbreakableTailLength = UnbreakableTailLength;
1713     if (Current->CanBreakBefore ||
1714         Current->isOneOf(tok::comment, tok::string_literal)) {
1715       UnbreakableTailLength = 0;
1716     } else {
1717       UnbreakableTailLength +=
1718           Current->ColumnWidth + Current->SpacesRequiredBefore;
1719     }
1720     Current = Current->Previous;
1721   }
1722 }
1723 
1724 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1725                                       const FormatToken &Tok,
1726                                       bool InFunctionDecl) {
1727   const FormatToken &Left = *Tok.Previous;
1728   const FormatToken &Right = Tok;
1729 
1730   if (Left.is(tok::semi))
1731     return 0;
1732 
1733   if (Style.Language == FormatStyle::LK_Java) {
1734     if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1735       return 1;
1736     if (Right.is(Keywords.kw_implements))
1737       return 2;
1738     if (Left.is(tok::comma) && Left.NestingLevel == 0)
1739       return 3;
1740   } else if (Style.Language == FormatStyle::LK_JavaScript) {
1741     if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1742       return 100;
1743     if (Left.is(TT_JsTypeColon))
1744       return 35;
1745   }
1746 
1747   if (Left.is(tok::comma) || (Right.is(tok::identifier) && Right.Next &&
1748                               Right.Next->is(TT_DictLiteral)))
1749     return 1;
1750   if (Right.is(tok::l_square)) {
1751     if (Style.Language == FormatStyle::LK_Proto)
1752       return 1;
1753     if (Left.is(tok::r_square))
1754       return 25;
1755     // Slightly prefer formatting local lambda definitions like functions.
1756     if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1757       return 35;
1758     if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1759                        TT_ArrayInitializerLSquare))
1760       return 500;
1761   }
1762 
1763   if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1764       Right.is(tok::kw_operator)) {
1765     if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1766       return 3;
1767     if (Left.is(TT_StartOfName))
1768       return 110;
1769     if (InFunctionDecl && Right.NestingLevel == 0)
1770       return Style.PenaltyReturnTypeOnItsOwnLine;
1771     return 200;
1772   }
1773   if (Right.is(TT_PointerOrReference))
1774     return 190;
1775   if (Right.is(TT_LambdaArrow))
1776     return 110;
1777   if (Left.is(tok::equal) && Right.is(tok::l_brace))
1778     return 150;
1779   if (Left.is(TT_CastRParen))
1780     return 100;
1781   if (Left.is(tok::coloncolon) ||
1782       (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1783     return 500;
1784   if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1785     return 5000;
1786 
1787   if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
1788     return 2;
1789 
1790   if (Right.isMemberAccess()) {
1791     // Breaking before the "./->" of a chained call/member access is reasonably
1792     // cheap, as formatting those with one call per line is generally
1793     // desirable. In particular, it should be cheaper to break before the call
1794     // than it is to break inside a call's parameters, which could lead to weird
1795     // "hanging" indents. The exception is the very last "./->" to support this
1796     // frequent pattern:
1797     //
1798     //   aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
1799     //       dddddddd);
1800     //
1801     // which might otherwise be blown up onto many lines. Here, clang-format
1802     // won't produce "hanging" indents anyway as there is no other trailing
1803     // call.
1804     //
1805     // Also apply higher penalty is not a call as that might lead to a wrapping
1806     // like:
1807     //
1808     //   aaaaaaa
1809     //       .aaaaaaaaa.bbbbbbbb(cccccccc);
1810     return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
1811                ? 150
1812                : 35;
1813   }
1814 
1815   if (Right.is(TT_TrailingAnnotation) &&
1816       (!Right.Next || Right.Next->isNot(tok::l_paren))) {
1817     // Moving trailing annotations to the next line is fine for ObjC method
1818     // declarations.
1819     if (Line.startsWith(TT_ObjCMethodSpecifier))
1820       return 10;
1821     // Generally, breaking before a trailing annotation is bad unless it is
1822     // function-like. It seems to be especially preferable to keep standard
1823     // annotations (i.e. "const", "final" and "override") on the same line.
1824     // Use a slightly higher penalty after ")" so that annotations like
1825     // "const override" are kept together.
1826     bool is_short_annotation = Right.TokenText.size() < 10;
1827     return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
1828   }
1829 
1830   // In for-loops, prefer breaking at ',' and ';'.
1831   if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
1832     return 4;
1833 
1834   // In Objective-C method expressions, prefer breaking before "param:" over
1835   // breaking after it.
1836   if (Right.is(TT_SelectorName))
1837     return 0;
1838   if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
1839     return Line.MightBeFunctionDecl ? 50 : 500;
1840 
1841   if (Left.is(tok::l_paren) && InFunctionDecl &&
1842       Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1843     return 100;
1844   if (Left.is(tok::l_paren) && Left.Previous &&
1845       Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
1846     return 1000;
1847   if (Left.is(tok::equal) && InFunctionDecl)
1848     return 110;
1849   if (Right.is(tok::r_brace))
1850     return 1;
1851   if (Left.is(TT_TemplateOpener))
1852     return 100;
1853   if (Left.opensScope()) {
1854     if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
1855       return 0;
1856     return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
1857                                    : 19;
1858   }
1859   if (Left.is(TT_JavaAnnotation))
1860     return 50;
1861 
1862   if (Right.is(tok::lessless)) {
1863     if (Left.is(tok::string_literal) &&
1864         (Right.NextOperator || Right.OperatorIndex != 1)) {
1865       StringRef Content = Left.TokenText;
1866       if (Content.startswith("\""))
1867         Content = Content.drop_front(1);
1868       if (Content.endswith("\""))
1869         Content = Content.drop_back(1);
1870       Content = Content.trim();
1871       if (Content.size() > 1 &&
1872           (Content.back() == ':' || Content.back() == '='))
1873         return 25;
1874     }
1875     return 1; // Breaking at a << is really cheap.
1876   }
1877   if (Left.is(TT_ConditionalExpr))
1878     return prec::Conditional;
1879   prec::Level Level = Left.getPrecedence();
1880   if (Level != prec::Unknown)
1881     return Level;
1882   Level = Right.getPrecedence();
1883   if (Level != prec::Unknown)
1884     return Level;
1885 
1886   return 3;
1887 }
1888 
1889 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
1890                                           const FormatToken &Left,
1891                                           const FormatToken &Right) {
1892   if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
1893     return true;
1894   if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
1895       Left.Tok.getObjCKeywordID() == tok::objc_property)
1896     return true;
1897   if (Right.is(tok::hashhash))
1898     return Left.is(tok::hash);
1899   if (Left.isOneOf(tok::hashhash, tok::hash))
1900     return Right.is(tok::hash);
1901   if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
1902     return Style.SpaceInEmptyParentheses;
1903   if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
1904     return (Right.is(TT_CastRParen) ||
1905             (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
1906                ? Style.SpacesInCStyleCastParentheses
1907                : Style.SpacesInParentheses;
1908   if (Right.isOneOf(tok::semi, tok::comma))
1909     return false;
1910   if (Right.is(tok::less) &&
1911       (Left.is(tok::kw_template) ||
1912        (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)))
1913     return true;
1914   if (Left.isOneOf(tok::exclaim, tok::tilde))
1915     return false;
1916   if (Left.is(tok::at) &&
1917       Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
1918                     tok::numeric_constant, tok::l_paren, tok::l_brace,
1919                     tok::kw_true, tok::kw_false))
1920     return false;
1921   if (Left.is(tok::colon))
1922     return !Left.is(TT_ObjCMethodExpr);
1923   if (Left.is(tok::coloncolon))
1924     return false;
1925   if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
1926     return false;
1927   if (Right.is(tok::ellipsis))
1928     return Left.Tok.isLiteral();
1929   if (Left.is(tok::l_square) && Right.is(tok::amp))
1930     return false;
1931   if (Right.is(TT_PointerOrReference))
1932     return (Left.is(tok::r_paren) && Left.MatchingParen &&
1933             (Left.MatchingParen->is(TT_OverloadedOperatorLParen) ||
1934              (Left.MatchingParen->Previous &&
1935               Left.MatchingParen->Previous->is(TT_FunctionDeclarationName)))) ||
1936            (Left.Tok.isLiteral() ||
1937             (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
1938              (Style.PointerAlignment != FormatStyle::PAS_Left ||
1939               Line.IsMultiVariableDeclStmt)));
1940   if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
1941       (!Left.is(TT_PointerOrReference) ||
1942        (Style.PointerAlignment != FormatStyle::PAS_Right &&
1943         !Line.IsMultiVariableDeclStmt)))
1944     return true;
1945   if (Left.is(TT_PointerOrReference))
1946     return Right.Tok.isLiteral() ||
1947            Right.isOneOf(TT_BlockComment, Keywords.kw_final,
1948                          Keywords.kw_override) ||
1949            (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
1950            (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
1951                            tok::l_paren) &&
1952             (Style.PointerAlignment != FormatStyle::PAS_Right &&
1953              !Line.IsMultiVariableDeclStmt) &&
1954             Left.Previous &&
1955             !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
1956   if (Right.is(tok::star) && Left.is(tok::l_paren))
1957     return false;
1958   if (Left.is(tok::l_square))
1959     return (Left.is(TT_ArrayInitializerLSquare) &&
1960             Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
1961            (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
1962             Right.isNot(tok::r_square));
1963   if (Right.is(tok::r_square))
1964     return Right.MatchingParen &&
1965            ((Style.SpacesInContainerLiterals &&
1966              Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
1967             (Style.SpacesInSquareBrackets &&
1968              Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
1969   if (Right.is(tok::l_square) &&
1970       !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
1971       !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
1972     return false;
1973   if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
1974     return !Left.Children.empty(); // No spaces in "{}".
1975   if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
1976       (Right.is(tok::r_brace) && Right.MatchingParen &&
1977        Right.MatchingParen->BlockKind != BK_Block))
1978     return !Style.Cpp11BracedListStyle;
1979   if (Left.is(TT_BlockComment))
1980     return !Left.TokenText.endswith("=*/");
1981   if (Right.is(tok::l_paren)) {
1982     if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
1983       return true;
1984     return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
1985            (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
1986             (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
1987                           tok::kw_switch, tok::kw_case, TT_ForEachMacro,
1988                           TT_ObjCForIn) ||
1989              (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
1990                            tok::kw_new, tok::kw_delete) &&
1991               (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
1992            (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
1993             (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
1994              Left.is(tok::r_paren)) &&
1995             Line.Type != LT_PreprocessorDirective);
1996   }
1997   if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
1998     return false;
1999   if (Right.is(TT_UnaryOperator))
2000     return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2001            (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2002   if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2003                     tok::r_paren) ||
2004        Left.isSimpleTypeSpecifier()) &&
2005       Right.is(tok::l_brace) && Right.getNextNonComment() &&
2006       Right.BlockKind != BK_Block)
2007     return false;
2008   if (Left.is(tok::period) || Right.is(tok::period))
2009     return false;
2010   if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2011     return false;
2012   if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2013       Left.MatchingParen->Previous &&
2014       Left.MatchingParen->Previous->is(tok::period))
2015     // A.<B>DoSomething();
2016     return false;
2017   if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2018     return false;
2019   return true;
2020 }
2021 
2022 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2023                                          const FormatToken &Right) {
2024   const FormatToken &Left = *Right.Previous;
2025   if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2026     return true; // Never ever merge two identifiers.
2027   if (Style.Language == FormatStyle::LK_Cpp) {
2028     if (Left.is(tok::kw_operator))
2029       return Right.is(tok::coloncolon);
2030   } else if (Style.Language == FormatStyle::LK_Proto) {
2031     if (Right.is(tok::period) &&
2032         Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2033                      Keywords.kw_repeated, Keywords.kw_extend))
2034       return true;
2035     if (Right.is(tok::l_paren) &&
2036         Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2037       return true;
2038   } else if (Style.Language == FormatStyle::LK_JavaScript) {
2039     if (Left.isOneOf(Keywords.kw_let, Keywords.kw_var, TT_JsFatArrow,
2040                      Keywords.kw_in, Keywords.kw_of))
2041       return true;
2042     if (Left.is(tok::kw_default) && Left.Previous &&
2043         Left.Previous->is(tok::kw_export))
2044       return true;
2045     if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2046       return true;
2047     if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2048       return false;
2049     if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2050         Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2051       return false;
2052     if (Left.is(tok::ellipsis))
2053       return false;
2054     if (Left.is(TT_TemplateCloser) &&
2055         !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2056                        Keywords.kw_implements, Keywords.kw_extends))
2057       // Type assertions ('<type>expr') are not followed by whitespace. Other
2058       // locations that should have whitespace following are identified by the
2059       // above set of follower tokens.
2060       return false;
2061   } else if (Style.Language == FormatStyle::LK_Java) {
2062     if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2063       return true;
2064     if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2065       return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2066     if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2067                       tok::kw_protected) ||
2068          Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2069                       Keywords.kw_native)) &&
2070         Right.is(TT_TemplateOpener))
2071       return true;
2072   }
2073   if (Left.is(TT_ImplicitStringLiteral))
2074     return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2075   if (Line.Type == LT_ObjCMethodDecl) {
2076     if (Left.is(TT_ObjCMethodSpecifier))
2077       return true;
2078     if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2079       // Don't space between ')' and <id>
2080       return false;
2081   }
2082   if (Line.Type == LT_ObjCProperty &&
2083       (Right.is(tok::equal) || Left.is(tok::equal)))
2084     return false;
2085 
2086   if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2087       Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2088     return true;
2089   if (Right.is(TT_OverloadedOperatorLParen))
2090     return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2091   if (Left.is(tok::comma))
2092     return true;
2093   if (Right.is(tok::comma))
2094     return false;
2095   if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2096     return true;
2097   if (Right.is(tok::colon)) {
2098     if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2099         !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2100       return false;
2101     if (Right.is(TT_ObjCMethodExpr))
2102       return false;
2103     if (Left.is(tok::question))
2104       return false;
2105     if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2106       return false;
2107     if (Right.is(TT_DictLiteral))
2108       return Style.SpacesInContainerLiterals;
2109     return true;
2110   }
2111   if (Left.is(TT_UnaryOperator))
2112     return Right.is(TT_BinaryOperator);
2113 
2114   // If the next token is a binary operator or a selector name, we have
2115   // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2116   if (Left.is(TT_CastRParen))
2117     return Style.SpaceAfterCStyleCast ||
2118            Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2119 
2120   if (Left.is(tok::greater) && Right.is(tok::greater))
2121     return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2122            (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2123   if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2124       Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2125     return false;
2126   if (!Style.SpaceBeforeAssignmentOperators &&
2127       Right.getPrecedence() == prec::Assignment)
2128     return false;
2129   if (Right.is(tok::coloncolon) && Left.isNot(tok::l_brace))
2130     return (Left.is(TT_TemplateOpener) &&
2131             Style.Standard == FormatStyle::LS_Cpp03) ||
2132            !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren,
2133                           tok::l_square) ||
2134              Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener));
2135   if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2136     return Style.SpacesInAngles;
2137   if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2138       (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2139        !Right.is(tok::r_paren)))
2140     return true;
2141   if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2142       Right.isNot(TT_FunctionTypeLParen))
2143     return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2144   if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2145       Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2146     return false;
2147   if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2148       Line.startsWith(tok::hash))
2149     return true;
2150   if (Right.is(TT_TrailingUnaryOperator))
2151     return false;
2152   if (Left.is(TT_RegexLiteral))
2153     return false;
2154   return spaceRequiredBetween(Line, Left, Right);
2155 }
2156 
2157 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2158 static bool isAllmanBrace(const FormatToken &Tok) {
2159   return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2160          !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2161 }
2162 
2163 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2164                                      const FormatToken &Right) {
2165   const FormatToken &Left = *Right.Previous;
2166   if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2167     return true;
2168 
2169   if (Style.Language == FormatStyle::LK_JavaScript) {
2170     // FIXME: This might apply to other languages and token kinds.
2171     if (Right.is(tok::char_constant) && Left.is(tok::plus) && Left.Previous &&
2172         Left.Previous->is(tok::char_constant))
2173       return true;
2174     if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2175         Left.Previous && Left.Previous->is(tok::equal) &&
2176         Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2177                             tok::kw_const) &&
2178         // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2179         // above.
2180         !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2181       // Object literals on the top level of a file are treated as "enum-style".
2182       // Each key/value pair is put on a separate line, instead of bin-packing.
2183       return true;
2184     if (Left.is(tok::l_brace) && Line.Level == 0 &&
2185         (Line.startsWith(tok::kw_enum) ||
2186          Line.startsWith(tok::kw_export, tok::kw_enum)))
2187       // JavaScript top-level enum key/value pairs are put on separate lines
2188       // instead of bin-packing.
2189       return true;
2190     if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2191         !Left.Children.empty())
2192       // Support AllowShortFunctionsOnASingleLine for JavaScript.
2193       return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2194              Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2195              (Left.NestingLevel == 0 && Line.Level == 0 &&
2196               Style.AllowShortFunctionsOnASingleLine ==
2197                   FormatStyle::SFS_Inline);
2198   } else if (Style.Language == FormatStyle::LK_Java) {
2199     if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2200         Right.Next->is(tok::string_literal))
2201       return true;
2202   }
2203 
2204   // If the last token before a '}' is a comma or a trailing comment, the
2205   // intention is to insert a line break after it in order to make shuffling
2206   // around entries easier.
2207   const FormatToken *BeforeClosingBrace = nullptr;
2208   if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
2209       Left.BlockKind != BK_Block && Left.MatchingParen)
2210     BeforeClosingBrace = Left.MatchingParen->Previous;
2211   else if (Right.MatchingParen &&
2212            Right.MatchingParen->isOneOf(tok::l_brace,
2213                                         TT_ArrayInitializerLSquare))
2214     BeforeClosingBrace = &Left;
2215   if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2216                              BeforeClosingBrace->isTrailingComment()))
2217     return true;
2218 
2219   if (Right.is(tok::comment))
2220     return Left.BlockKind != BK_BracedInit &&
2221            Left.isNot(TT_CtorInitializerColon) &&
2222            (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2223   if (Left.isTrailingComment())
2224     return true;
2225   if (Left.isStringLiteral() &&
2226       (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral)))
2227     return true;
2228   if (Right.Previous->IsUnterminatedLiteral)
2229     return true;
2230   if (Right.is(tok::lessless) && Right.Next &&
2231       Right.Previous->is(tok::string_literal) &&
2232       Right.Next->is(tok::string_literal))
2233     return true;
2234   if (Right.Previous->ClosesTemplateDeclaration &&
2235       Right.Previous->MatchingParen &&
2236       Right.Previous->MatchingParen->NestingLevel == 0 &&
2237       Style.AlwaysBreakTemplateDeclarations)
2238     return true;
2239   if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2240       Style.BreakConstructorInitializersBeforeComma &&
2241       !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2242     return true;
2243   if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2244     // Raw string literals are special wrt. line breaks. The author has made a
2245     // deliberate choice and might have aligned the contents of the string
2246     // literal accordingly. Thus, we try keep existing line breaks.
2247     return Right.NewlinesBefore > 0;
2248   if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2249       Style.Language == FormatStyle::LK_Proto)
2250     // Don't put enums onto single lines in protocol buffers.
2251     return true;
2252   if (Right.is(TT_InlineASMBrace))
2253     return Right.HasUnescapedNewline;
2254   if (isAllmanBrace(Left) || isAllmanBrace(Right))
2255     return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2256            (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2257            (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2258   if (Style.Language == FormatStyle::LK_Proto && Left.isNot(tok::l_brace) &&
2259       Right.is(TT_SelectorName))
2260     return true;
2261   if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2262     return true;
2263 
2264   if ((Style.Language == FormatStyle::LK_Java ||
2265        Style.Language == FormatStyle::LK_JavaScript) &&
2266       Left.is(TT_LeadingJavaAnnotation) &&
2267       Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2268       (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2269     return true;
2270 
2271   return false;
2272 }
2273 
2274 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2275                                     const FormatToken &Right) {
2276   const FormatToken &Left = *Right.Previous;
2277 
2278   // Language-specific stuff.
2279   if (Style.Language == FormatStyle::LK_Java) {
2280     if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2281                      Keywords.kw_implements))
2282       return false;
2283     if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2284                       Keywords.kw_implements))
2285       return true;
2286   } else if (Style.Language == FormatStyle::LK_JavaScript) {
2287     if (Left.is(tok::kw_return))
2288       return false; // Otherwise a semicolon is inserted.
2289     if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2290       return false;
2291     if (Left.is(TT_JsTypeColon))
2292       return true;
2293     if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2294       return false;
2295     if (Left.is(Keywords.kw_in))
2296       return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
2297     if (Right.is(Keywords.kw_in))
2298       return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
2299   }
2300 
2301   if (Left.is(tok::at))
2302     return false;
2303   if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2304     return false;
2305   if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2306     return !Right.is(tok::l_paren);
2307   if (Right.is(TT_PointerOrReference))
2308     return Line.IsMultiVariableDeclStmt ||
2309            (Style.PointerAlignment == FormatStyle::PAS_Right &&
2310             (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2311   if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2312       Right.is(tok::kw_operator))
2313     return true;
2314   if (Left.is(TT_PointerOrReference))
2315     return false;
2316   if (Right.isTrailingComment())
2317     // We rely on MustBreakBefore being set correctly here as we should not
2318     // change the "binding" behavior of a comment.
2319     // The first comment in a braced lists is always interpreted as belonging to
2320     // the first list element. Otherwise, it should be placed outside of the
2321     // list.
2322     return Left.BlockKind == BK_BracedInit;
2323   if (Left.is(tok::question) && Right.is(tok::colon))
2324     return false;
2325   if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2326     return Style.BreakBeforeTernaryOperators;
2327   if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2328     return !Style.BreakBeforeTernaryOperators;
2329   if (Right.is(TT_InheritanceColon))
2330     return true;
2331   if (Right.is(tok::colon) &&
2332       !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2333     return false;
2334   if (Left.is(tok::colon) && (Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))
2335     return true;
2336   if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2337                                     Right.Next->is(TT_ObjCMethodExpr)))
2338     return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2339   if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2340     return true;
2341   if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2342     return true;
2343   if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2344                     TT_OverloadedOperator))
2345     return false;
2346   if (Left.is(TT_RangeBasedForLoopColon))
2347     return true;
2348   if (Right.is(TT_RangeBasedForLoopColon))
2349     return false;
2350   if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2351       Left.is(tok::kw_operator))
2352     return false;
2353   if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2354       Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2355     return false;
2356   if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2357     return false;
2358   if (Left.is(tok::l_paren) && Left.Previous &&
2359       (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2360     return false;
2361   if (Right.is(TT_ImplicitStringLiteral))
2362     return false;
2363 
2364   if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2365     return false;
2366   if (Right.is(tok::r_square) && Right.MatchingParen &&
2367       Right.MatchingParen->is(TT_LambdaLSquare))
2368     return false;
2369 
2370   // We only break before r_brace if there was a corresponding break before
2371   // the l_brace, which is tracked by BreakBeforeClosingBrace.
2372   if (Right.is(tok::r_brace))
2373     return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2374 
2375   // Allow breaking after a trailing annotation, e.g. after a method
2376   // declaration.
2377   if (Left.is(TT_TrailingAnnotation))
2378     return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2379                           tok::less, tok::coloncolon);
2380 
2381   if (Right.is(tok::kw___attribute))
2382     return true;
2383 
2384   if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2385     return true;
2386 
2387   if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2388     return true;
2389 
2390   if (Left.is(TT_CtorInitializerComma) &&
2391       Style.BreakConstructorInitializersBeforeComma)
2392     return false;
2393   if (Right.is(TT_CtorInitializerComma) &&
2394       Style.BreakConstructorInitializersBeforeComma)
2395     return true;
2396   if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2397       (Left.is(tok::less) && Right.is(tok::less)))
2398     return false;
2399   if (Right.is(TT_BinaryOperator) &&
2400       Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2401       (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2402        Right.getPrecedence() != prec::Assignment))
2403     return true;
2404   if (Left.is(TT_ArrayInitializerLSquare))
2405     return true;
2406   if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2407     return true;
2408   if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2409       !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2410       Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2411       (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2412        Left.getPrecedence() == prec::Assignment))
2413     return true;
2414   return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2415                       tok::kw_class, tok::kw_struct) ||
2416          Right.isMemberAccess() ||
2417          Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2418                        tok::colon, tok::l_square, tok::at) ||
2419          (Left.is(tok::r_paren) &&
2420           Right.isOneOf(tok::identifier, tok::kw_const)) ||
2421          (Left.is(tok::l_paren) && !Right.is(tok::r_paren));
2422 }
2423 
2424 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2425   llvm::errs() << "AnnotatedTokens:\n";
2426   const FormatToken *Tok = Line.First;
2427   while (Tok) {
2428     llvm::errs() << " M=" << Tok->MustBreakBefore
2429                  << " C=" << Tok->CanBreakBefore
2430                  << " T=" << getTokenTypeName(Tok->Type)
2431                  << " S=" << Tok->SpacesRequiredBefore
2432                  << " B=" << Tok->BlockParameterCount
2433                  << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2434                  << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2435                  << " FakeLParens=";
2436     for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2437       llvm::errs() << Tok->FakeLParens[i] << "/";
2438     llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n";
2439     if (!Tok->Next)
2440       assert(Tok == Line.Last);
2441     Tok = Tok->Next;
2442   }
2443   llvm::errs() << "----\n";
2444 }
2445 
2446 } // namespace format
2447 } // namespace clang
2448