• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/Support/Debug.h"
19 
20 namespace clang {
21 namespace format {
22 
23 namespace {
24 
25 /// \brief A parser that gathers additional information about tokens.
26 ///
27 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
28 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
29 /// into template parameter lists.
30 class AnnotatingParser {
31 public:
AnnotatingParser(AnnotatedLine & Line,IdentifierInfo & Ident_in)32   AnnotatingParser(AnnotatedLine &Line, IdentifierInfo &Ident_in)
33       : Line(Line), CurrentToken(Line.First), KeywordVirtualFound(false),
34         NameFound(false), AutoFound(false), Ident_in(Ident_in) {
35     Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
36   }
37 
38 private:
parseAngle()39   bool parseAngle() {
40     if (CurrentToken == NULL)
41       return false;
42     ScopedContextCreator ContextCreator(*this, tok::less, 10);
43     FormatToken *Left = CurrentToken->Previous;
44     Contexts.back().IsExpression = false;
45     while (CurrentToken != NULL) {
46       if (CurrentToken->is(tok::greater)) {
47         Left->MatchingParen = CurrentToken;
48         CurrentToken->MatchingParen = Left;
49         CurrentToken->Type = TT_TemplateCloser;
50         next();
51         return true;
52       }
53       if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace,
54                                 tok::question, tok::colon))
55         return false;
56       // If a && or || is found and interpreted as a binary operator, this set
57       // of angles is likely part of something like "a < b && c > d". If the
58       // angles are inside an expression, the ||/&& might also be a binary
59       // operator that was misinterpreted because we are parsing template
60       // parameters.
61       // FIXME: This is getting out of hand, write a decent parser.
62       if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
63           (CurrentToken->Previous->Type == TT_BinaryOperator ||
64            Contexts[Contexts.size() - 2].IsExpression) &&
65           Line.First->isNot(tok::kw_template))
66         return false;
67       updateParameterCount(Left, CurrentToken);
68       if (!consumeToken())
69         return false;
70     }
71     return false;
72   }
73 
parseParens(bool LookForDecls=false)74   bool parseParens(bool LookForDecls = false) {
75     if (CurrentToken == NULL)
76       return false;
77     ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
78 
79     // FIXME: This is a bit of a hack. Do better.
80     Contexts.back().ColonIsForRangeExpr =
81         Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
82 
83     bool StartsObjCMethodExpr = false;
84     FormatToken *Left = CurrentToken->Previous;
85     if (CurrentToken->is(tok::caret)) {
86       // ^( starts a block.
87       Left->Type = TT_ObjCBlockLParen;
88     } else if (FormatToken *MaybeSel = Left->Previous) {
89       // @selector( starts a selector.
90       if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
91           MaybeSel->Previous->is(tok::at)) {
92         StartsObjCMethodExpr = true;
93       }
94     }
95 
96     if (Left->Previous && Left->Previous->is(tok::kw_static_assert))
97       Contexts.back().IsExpression = true;
98 
99     if (StartsObjCMethodExpr) {
100       Contexts.back().ColonIsObjCMethodExpr = true;
101       Left->Type = TT_ObjCMethodExpr;
102     }
103 
104     bool MightBeFunctionType = CurrentToken->is(tok::star);
105     bool HasMultipleLines = false;
106     bool HasMultipleParametersOnALine = false;
107     while (CurrentToken != NULL) {
108       // LookForDecls is set when "if (" has been seen. Check for
109       // 'identifier' '*' 'identifier' followed by not '=' -- this
110       // '*' has to be a binary operator but determineStarAmpUsage() will
111       // categorize it as an unary operator, so set the right type here.
112       if (LookForDecls && CurrentToken->Next) {
113         FormatToken *Prev = CurrentToken->getPreviousNonComment();
114         if (Prev) {
115           FormatToken *PrevPrev = Prev->getPreviousNonComment();
116           FormatToken *Next = CurrentToken->Next;
117           if (PrevPrev && PrevPrev->is(tok::identifier) &&
118               Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
119               CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
120             Prev->Type = TT_BinaryOperator;
121             LookForDecls = false;
122           }
123         }
124       }
125 
126       if (CurrentToken->is(tok::r_paren)) {
127         if (MightBeFunctionType && CurrentToken->Next &&
128             (CurrentToken->Next->is(tok::l_paren) ||
129              (CurrentToken->Next->is(tok::l_square) &&
130               !Contexts.back().IsExpression)))
131           Left->Type = TT_FunctionTypeLParen;
132         Left->MatchingParen = CurrentToken;
133         CurrentToken->MatchingParen = Left;
134 
135         if (StartsObjCMethodExpr) {
136           CurrentToken->Type = TT_ObjCMethodExpr;
137           if (Contexts.back().FirstObjCSelectorName != NULL) {
138             Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
139                 Contexts.back().LongestObjCSelectorName;
140           }
141         }
142 
143         if (!HasMultipleLines)
144           Left->PackingKind = PPK_Inconclusive;
145         else if (HasMultipleParametersOnALine)
146           Left->PackingKind = PPK_BinPacked;
147         else
148           Left->PackingKind = PPK_OnePerLine;
149 
150         next();
151         return true;
152       }
153       if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
154         return false;
155       if (CurrentToken->Previous->Type == TT_PointerOrReference &&
156           CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
157                                                     tok::coloncolon))
158         MightBeFunctionType = true;
159       updateParameterCount(Left, CurrentToken);
160       if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
161           !CurrentToken->Next->HasUnescapedNewline &&
162           !CurrentToken->Next->isTrailingComment())
163         HasMultipleParametersOnALine = true;
164       if (!consumeToken())
165         return false;
166       if (CurrentToken && CurrentToken->HasUnescapedNewline)
167         HasMultipleLines = true;
168     }
169     return false;
170   }
171 
parseSquare()172   bool parseSquare() {
173     if (!CurrentToken)
174       return false;
175 
176     // A '[' could be an index subscript (after an identifier or after
177     // ')' or ']'), it could be the start of an Objective-C method
178     // expression, or it could the the start of an Objective-C array literal.
179     FormatToken *Left = CurrentToken->Previous;
180     FormatToken *Parent = Left->getPreviousNonComment();
181     bool StartsObjCMethodExpr =
182         Contexts.back().CanBeExpression &&
183         (!Parent || Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
184                                     tok::kw_return, tok::kw_throw) ||
185          Parent->isUnaryOperator() || Parent->Type == TT_ObjCForIn ||
186          Parent->Type == TT_CastRParen ||
187          getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
188     ScopedContextCreator ContextCreator(*this, tok::l_square, 10);
189     Contexts.back().IsExpression = true;
190     bool StartsObjCArrayLiteral = Parent && Parent->is(tok::at);
191 
192     if (StartsObjCMethodExpr) {
193       Contexts.back().ColonIsObjCMethodExpr = true;
194       Left->Type = TT_ObjCMethodExpr;
195     } else if (StartsObjCArrayLiteral) {
196       Left->Type = TT_ObjCArrayLiteral;
197     }
198 
199     while (CurrentToken != NULL) {
200       if (CurrentToken->is(tok::r_square)) {
201         if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren)) {
202           // An ObjC method call is rarely followed by an open parenthesis.
203           // FIXME: Do we incorrectly label ":" with this?
204           StartsObjCMethodExpr = false;
205           Left->Type = TT_Unknown;
206         }
207         if (StartsObjCMethodExpr) {
208           CurrentToken->Type = TT_ObjCMethodExpr;
209           // determineStarAmpUsage() thinks that '*' '[' is allocating an
210           // array of pointers, but if '[' starts a selector then '*' is a
211           // binary operator.
212           if (Parent != NULL && Parent->Type == TT_PointerOrReference)
213             Parent->Type = TT_BinaryOperator;
214         } else if (StartsObjCArrayLiteral) {
215           CurrentToken->Type = TT_ObjCArrayLiteral;
216         }
217         Left->MatchingParen = CurrentToken;
218         CurrentToken->MatchingParen = Left;
219         if (Contexts.back().FirstObjCSelectorName != NULL)
220           Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
221               Contexts.back().LongestObjCSelectorName;
222         next();
223         return true;
224       }
225       if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
226         return false;
227       updateParameterCount(Left, CurrentToken);
228       if (!consumeToken())
229         return false;
230     }
231     return false;
232   }
233 
parseBrace()234   bool parseBrace() {
235     if (CurrentToken != NULL) {
236       ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
237       FormatToken *Left = CurrentToken->Previous;
238 
239       FormatToken *Parent = Left->getPreviousNonComment();
240       bool StartsObjCDictLiteral = Parent && Parent->is(tok::at);
241       if (StartsObjCDictLiteral) {
242         Contexts.back().ColonIsObjCDictLiteral = true;
243         Left->Type = TT_ObjCDictLiteral;
244       }
245 
246       while (CurrentToken != NULL) {
247         if (CurrentToken->is(tok::r_brace)) {
248           if (StartsObjCDictLiteral)
249             CurrentToken->Type = TT_ObjCDictLiteral;
250           Left->MatchingParen = CurrentToken;
251           CurrentToken->MatchingParen = Left;
252           next();
253           return true;
254         }
255         if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
256           return false;
257         updateParameterCount(Left, CurrentToken);
258         if (!consumeToken())
259           return false;
260       }
261     }
262     // No closing "}" found, this probably starts a definition.
263     Line.StartsDefinition = true;
264     return true;
265   }
266 
updateParameterCount(FormatToken * Left,FormatToken * Current)267   void updateParameterCount(FormatToken *Left, FormatToken *Current) {
268     if (Current->is(tok::comma)) {
269       ++Left->ParameterCount;
270     } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
271       Left->ParameterCount = 1;
272     }
273   }
274 
parseConditional()275   bool parseConditional() {
276     while (CurrentToken != NULL) {
277       if (CurrentToken->is(tok::colon)) {
278         CurrentToken->Type = TT_ConditionalExpr;
279         next();
280         return true;
281       }
282       if (!consumeToken())
283         return false;
284     }
285     return false;
286   }
287 
parseTemplateDeclaration()288   bool parseTemplateDeclaration() {
289     if (CurrentToken != NULL && CurrentToken->is(tok::less)) {
290       CurrentToken->Type = TT_TemplateOpener;
291       next();
292       if (!parseAngle())
293         return false;
294       if (CurrentToken != NULL)
295         CurrentToken->Previous->ClosesTemplateDeclaration = true;
296       return true;
297     }
298     return false;
299   }
300 
consumeToken()301   bool consumeToken() {
302     FormatToken *Tok = CurrentToken;
303     next();
304     switch (Tok->Tok.getKind()) {
305     case tok::plus:
306     case tok::minus:
307       if (Tok->Previous == NULL && Line.MustBeDeclaration)
308         Tok->Type = TT_ObjCMethodSpecifier;
309       break;
310     case tok::colon:
311       if (Tok->Previous == NULL)
312         return false;
313       // Colons from ?: are handled in parseConditional().
314       if (Tok->Previous->is(tok::r_paren) && Contexts.size() == 1) {
315         Tok->Type = TT_CtorInitializerColon;
316       } else if (Contexts.back().ColonIsObjCDictLiteral) {
317         Tok->Type = TT_ObjCDictLiteral;
318       } else if (Contexts.back().ColonIsObjCMethodExpr ||
319                  Line.First->Type == TT_ObjCMethodSpecifier) {
320         Tok->Type = TT_ObjCMethodExpr;
321         Tok->Previous->Type = TT_ObjCSelectorName;
322         if (Tok->Previous->CodePointCount >
323             Contexts.back().LongestObjCSelectorName) {
324           Contexts.back().LongestObjCSelectorName =
325               Tok->Previous->CodePointCount;
326         }
327         if (Contexts.back().FirstObjCSelectorName == NULL)
328           Contexts.back().FirstObjCSelectorName = Tok->Previous;
329       } else if (Contexts.back().ColonIsForRangeExpr) {
330         Tok->Type = TT_RangeBasedForLoopColon;
331       } else if (Contexts.size() == 1) {
332         Tok->Type = TT_InheritanceColon;
333       } else if (Contexts.back().ContextKind == tok::l_paren) {
334         Tok->Type = TT_InlineASMColon;
335       }
336       break;
337     case tok::kw_if:
338     case tok::kw_while:
339       if (CurrentToken != NULL && CurrentToken->is(tok::l_paren)) {
340         next();
341         if (!parseParens(/*LookForDecls=*/true))
342           return false;
343       }
344       break;
345     case tok::kw_for:
346       Contexts.back().ColonIsForRangeExpr = true;
347       next();
348       if (!parseParens())
349         return false;
350       break;
351     case tok::l_paren:
352       if (!parseParens())
353         return false;
354       if (Line.MustBeDeclaration && NameFound && !Contexts.back().IsExpression)
355         Line.MightBeFunctionDecl = true;
356       break;
357     case tok::l_square:
358       if (!parseSquare())
359         return false;
360       break;
361     case tok::l_brace:
362       if (!parseBrace())
363         return false;
364       break;
365     case tok::less:
366       if (Tok->Previous && !Tok->Previous->Tok.isLiteral() && parseAngle())
367         Tok->Type = TT_TemplateOpener;
368       else {
369         Tok->Type = TT_BinaryOperator;
370         CurrentToken = Tok;
371         next();
372       }
373       break;
374     case tok::r_paren:
375     case tok::r_square:
376       return false;
377     case tok::r_brace:
378       // Lines can start with '}'.
379       if (Tok->Previous != NULL)
380         return false;
381       break;
382     case tok::greater:
383       Tok->Type = TT_BinaryOperator;
384       break;
385     case tok::kw_operator:
386       while (CurrentToken && CurrentToken->isNot(tok::l_paren)) {
387         if (CurrentToken->isOneOf(tok::star, tok::amp))
388           CurrentToken->Type = TT_PointerOrReference;
389         consumeToken();
390       }
391       if (CurrentToken) {
392         CurrentToken->Type = TT_OverloadedOperatorLParen;
393         if (CurrentToken->Previous->Type == TT_BinaryOperator)
394           CurrentToken->Previous->Type = TT_OverloadedOperator;
395       }
396       break;
397     case tok::question:
398       parseConditional();
399       break;
400     case tok::kw_template:
401       parseTemplateDeclaration();
402       break;
403     case tok::identifier:
404       if (Line.First->is(tok::kw_for) &&
405           Tok->Tok.getIdentifierInfo() == &Ident_in)
406         Tok->Type = TT_ObjCForIn;
407       break;
408     case tok::comma:
409       if (Contexts.back().FirstStartOfName)
410         Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
411       if (Contexts.back().InCtorInitializer)
412         Tok->Type = TT_CtorInitializerComma;
413       break;
414     default:
415       break;
416     }
417     return true;
418   }
419 
parseIncludeDirective()420   void parseIncludeDirective() {
421     next();
422     if (CurrentToken != NULL && CurrentToken->is(tok::less)) {
423       next();
424       while (CurrentToken != NULL) {
425         if (CurrentToken->isNot(tok::comment) || CurrentToken->Next)
426           CurrentToken->Type = TT_ImplicitStringLiteral;
427         next();
428       }
429     } else {
430       while (CurrentToken != NULL) {
431         if (CurrentToken->is(tok::string_literal))
432           // Mark these string literals as "implicit" literals, too, so that
433           // they are not split or line-wrapped.
434           CurrentToken->Type = TT_ImplicitStringLiteral;
435         next();
436       }
437     }
438   }
439 
parseWarningOrError()440   void parseWarningOrError() {
441     next();
442     // We still want to format the whitespace left of the first token of the
443     // warning or error.
444     next();
445     while (CurrentToken != NULL) {
446       CurrentToken->Type = TT_ImplicitStringLiteral;
447       next();
448     }
449   }
450 
parsePreprocessorDirective()451   void parsePreprocessorDirective() {
452     next();
453     if (CurrentToken == NULL)
454       return;
455     // Hashes in the middle of a line can lead to any strange token
456     // sequence.
457     if (CurrentToken->Tok.getIdentifierInfo() == NULL)
458       return;
459     switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
460     case tok::pp_include:
461     case tok::pp_import:
462       parseIncludeDirective();
463       break;
464     case tok::pp_error:
465     case tok::pp_warning:
466       parseWarningOrError();
467       break;
468     case tok::pp_if:
469     case tok::pp_elif:
470       parseLine();
471       break;
472     default:
473       break;
474     }
475     while (CurrentToken != NULL)
476       next();
477   }
478 
479 public:
parseLine()480   LineType parseLine() {
481     int PeriodsAndArrows = 0;
482     FormatToken *LastPeriodOrArrow = NULL;
483     bool CanBeBuilderTypeStmt = true;
484     if (CurrentToken->is(tok::hash)) {
485       parsePreprocessorDirective();
486       return LT_PreprocessorDirective;
487     }
488     while (CurrentToken != NULL) {
489       if (CurrentToken->is(tok::kw_virtual))
490         KeywordVirtualFound = true;
491       if (CurrentToken->isOneOf(tok::period, tok::arrow)) {
492         ++PeriodsAndArrows;
493         LastPeriodOrArrow = CurrentToken;
494       }
495       FormatToken *TheToken = CurrentToken;
496       if (!consumeToken())
497         return LT_Invalid;
498       if (TheToken->getPrecedence() > prec::Assignment &&
499           TheToken->Type == TT_BinaryOperator)
500         CanBeBuilderTypeStmt = false;
501     }
502     if (KeywordVirtualFound)
503       return LT_VirtualFunctionDecl;
504 
505     // Assume a builder-type call if there are 2 or more "." and "->".
506     if (PeriodsAndArrows >= 2 && CanBeBuilderTypeStmt) {
507       LastPeriodOrArrow->LastInChainOfCalls = true;
508       return LT_BuilderTypeCall;
509     }
510 
511     if (Line.First->Type == TT_ObjCMethodSpecifier) {
512       if (Contexts.back().FirstObjCSelectorName != NULL)
513         Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
514             Contexts.back().LongestObjCSelectorName;
515       return LT_ObjCMethodDecl;
516     }
517 
518     return LT_Other;
519   }
520 
521 private:
next()522   void next() {
523     if (CurrentToken != NULL) {
524       determineTokenType(*CurrentToken);
525       CurrentToken->BindingStrength = Contexts.back().BindingStrength;
526     }
527 
528     if (CurrentToken != NULL)
529       CurrentToken = CurrentToken->Next;
530 
531     // Reset token type in case we have already looked at it and then recovered
532     // from an error (e.g. failure to find the matching >).
533     if (CurrentToken != NULL)
534       CurrentToken->Type = TT_Unknown;
535   }
536 
537   /// \brief A struct to hold information valid in a specific context, e.g.
538   /// a pair of parenthesis.
539   struct Context {
Contextclang::format::__anon157a22350111::AnnotatingParser::Context540     Context(tok::TokenKind ContextKind, unsigned BindingStrength,
541             bool IsExpression)
542         : ContextKind(ContextKind), BindingStrength(BindingStrength),
543           LongestObjCSelectorName(0), ColonIsForRangeExpr(false),
544           ColonIsObjCDictLiteral(false), ColonIsObjCMethodExpr(false),
545           FirstObjCSelectorName(NULL), FirstStartOfName(NULL),
546           IsExpression(IsExpression), CanBeExpression(true),
547           InCtorInitializer(false) {}
548 
549     tok::TokenKind ContextKind;
550     unsigned BindingStrength;
551     unsigned LongestObjCSelectorName;
552     bool ColonIsForRangeExpr;
553     bool ColonIsObjCDictLiteral;
554     bool ColonIsObjCMethodExpr;
555     FormatToken *FirstObjCSelectorName;
556     FormatToken *FirstStartOfName;
557     bool IsExpression;
558     bool CanBeExpression;
559     bool InCtorInitializer;
560   };
561 
562   /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
563   /// of each instance.
564   struct ScopedContextCreator {
565     AnnotatingParser &P;
566 
ScopedContextCreatorclang::format::__anon157a22350111::AnnotatingParser::ScopedContextCreator567     ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
568                          unsigned Increase)
569         : P(P) {
570       P.Contexts.push_back(Context(ContextKind,
571                                    P.Contexts.back().BindingStrength + Increase,
572                                    P.Contexts.back().IsExpression));
573     }
574 
~ScopedContextCreatorclang::format::__anon157a22350111::AnnotatingParser::ScopedContextCreator575     ~ScopedContextCreator() { P.Contexts.pop_back(); }
576   };
577 
determineTokenType(FormatToken & Current)578   void determineTokenType(FormatToken &Current) {
579     if (Current.getPrecedence() == prec::Assignment &&
580         (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
581       Contexts.back().IsExpression = true;
582       for (FormatToken *Previous = Current.Previous;
583            Previous && Previous->isNot(tok::comma);
584            Previous = Previous->Previous) {
585         if (Previous->is(tok::r_square))
586           Previous = Previous->MatchingParen;
587         if (Previous->Type == TT_BinaryOperator &&
588             Previous->isOneOf(tok::star, tok::amp)) {
589           Previous->Type = TT_PointerOrReference;
590         }
591       }
592     } else if (Current.isOneOf(tok::kw_return, tok::kw_throw) ||
593                (Current.is(tok::l_paren) && !Line.MustBeDeclaration &&
594                 !Line.InPPDirective &&
595                 (!Current.Previous || Current.Previous->isNot(tok::kw_for)))) {
596       Contexts.back().IsExpression = true;
597     } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
598       for (FormatToken *Previous = Current.Previous;
599            Previous && Previous->isOneOf(tok::star, tok::amp);
600            Previous = Previous->Previous)
601         Previous->Type = TT_PointerOrReference;
602     } else if (Current.Previous &&
603                Current.Previous->Type == TT_CtorInitializerColon) {
604       Contexts.back().IsExpression = true;
605       Contexts.back().InCtorInitializer = true;
606     } else if (Current.is(tok::kw_new)) {
607       Contexts.back().CanBeExpression = false;
608     } else if (Current.is(tok::semi)) {
609       // This should be the condition or increment in a for-loop.
610       Contexts.back().IsExpression = true;
611     }
612 
613     if (Current.Type == TT_Unknown) {
614       if (isStartOfName(Current)) {
615         Contexts.back().FirstStartOfName = &Current;
616         Current.Type = TT_StartOfName;
617         NameFound = true;
618       } else if (Current.is(tok::kw_auto)) {
619         AutoFound = true;
620       } else if (Current.is(tok::arrow) && AutoFound &&
621                  Line.MustBeDeclaration) {
622         Current.Type = TT_TrailingReturnArrow;
623       } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
624         Current.Type =
625             determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
626                                                Contexts.back().IsExpression);
627       } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
628         Current.Type = determinePlusMinusCaretUsage(Current);
629       } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
630         Current.Type = determineIncrementUsage(Current);
631       } else if (Current.is(tok::exclaim)) {
632         Current.Type = TT_UnaryOperator;
633       } else if (Current.isBinaryOperator()) {
634         Current.Type = TT_BinaryOperator;
635       } else if (Current.is(tok::comment)) {
636         if (Current.TokenText.startswith("//"))
637           Current.Type = TT_LineComment;
638         else
639           Current.Type = TT_BlockComment;
640       } else if (Current.is(tok::r_paren)) {
641         FormatToken *LeftOfParens = NULL;
642         if (Current.MatchingParen)
643           LeftOfParens = Current.MatchingParen->getPreviousNonComment();
644         bool IsCast = false;
645         bool ParensAreEmpty = Current.Previous == Current.MatchingParen;
646         bool ParensAreType = !Current.Previous ||
647                              Current.Previous->Type == TT_PointerOrReference ||
648                              Current.Previous->Type == TT_TemplateCloser ||
649                              isSimpleTypeSpecifier(*Current.Previous);
650         bool ParensCouldEndDecl =
651             Current.Next &&
652             Current.Next->isOneOf(tok::equal, tok::semi, tok::l_brace);
653         bool IsSizeOfOrAlignOf =
654             LeftOfParens &&
655             LeftOfParens->isOneOf(tok::kw_sizeof, tok::kw_alignof);
656         if (ParensAreType && !ParensCouldEndDecl && !IsSizeOfOrAlignOf &&
657             (Contexts.back().IsExpression ||
658              (Current.Next && Current.Next->isBinaryOperator())))
659           IsCast = true;
660         if (Current.Next && Current.Next->isNot(tok::string_literal) &&
661             (Current.Next->Tok.isLiteral() ||
662              Current.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
663           IsCast = true;
664         // If there is an identifier after the (), it is likely a cast, unless
665         // there is also an identifier before the ().
666         if (LeftOfParens && (LeftOfParens->Tok.getIdentifierInfo() == NULL ||
667                              LeftOfParens->is(tok::kw_return)) &&
668             LeftOfParens->Type != TT_OverloadedOperator &&
669             LeftOfParens->Type != TT_TemplateCloser && Current.Next &&
670             Current.Next->is(tok::identifier))
671           IsCast = true;
672         if (IsCast && !ParensAreEmpty)
673           Current.Type = TT_CastRParen;
674       } else if (Current.is(tok::at) && Current.Next) {
675         switch (Current.Next->Tok.getObjCKeywordID()) {
676         case tok::objc_interface:
677         case tok::objc_implementation:
678         case tok::objc_protocol:
679           Current.Type = TT_ObjCDecl;
680           break;
681         case tok::objc_property:
682           Current.Type = TT_ObjCProperty;
683           break;
684         default:
685           break;
686         }
687       } else if (Current.is(tok::period)) {
688         FormatToken *PreviousNoComment = Current.getPreviousNonComment();
689         if (PreviousNoComment &&
690             PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
691           Current.Type = TT_DesignatedInitializerPeriod;
692       }
693     }
694   }
695 
696   /// \brief Take a guess at whether \p Tok starts a name of a function or
697   /// variable declaration.
698   ///
699   /// This is a heuristic based on whether \p Tok is an identifier following
700   /// something that is likely a type.
isStartOfName(const FormatToken & Tok)701   bool isStartOfName(const FormatToken &Tok) {
702     if (Tok.isNot(tok::identifier) || Tok.Previous == NULL)
703       return false;
704 
705     // Skip "const" as it does not have an influence on whether this is a name.
706     FormatToken *PreviousNotConst = Tok.Previous;
707     while (PreviousNotConst != NULL && PreviousNotConst->is(tok::kw_const))
708       PreviousNotConst = PreviousNotConst->Previous;
709 
710     if (PreviousNotConst == NULL)
711       return false;
712 
713     bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
714                        PreviousNotConst->Previous &&
715                        PreviousNotConst->Previous->is(tok::hash);
716 
717     return (!IsPPKeyword && PreviousNotConst->is(tok::identifier)) ||
718            PreviousNotConst->Type == TT_PointerOrReference ||
719            PreviousNotConst->Type == TT_TemplateCloser ||
720            isSimpleTypeSpecifier(*PreviousNotConst);
721   }
722 
723   /// \brief Return the type of the given token assuming it is * or &.
determineStarAmpUsage(const FormatToken & Tok,bool IsExpression)724   TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression) {
725     const FormatToken *PrevToken = Tok.getPreviousNonComment();
726     if (PrevToken == NULL)
727       return TT_UnaryOperator;
728 
729     const FormatToken *NextToken = Tok.getNextNonComment();
730     if (NextToken == NULL)
731       return TT_Unknown;
732 
733     if (PrevToken->is(tok::coloncolon) ||
734         (PrevToken->is(tok::l_paren) && !IsExpression))
735       return TT_PointerOrReference;
736 
737     if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
738                            tok::comma, tok::semi, tok::kw_return, tok::colon,
739                            tok::equal, tok::kw_delete) ||
740         PrevToken->Type == TT_BinaryOperator ||
741         PrevToken->Type == TT_UnaryOperator || PrevToken->Type == TT_CastRParen)
742       return TT_UnaryOperator;
743 
744     if (NextToken->is(tok::l_square))
745       return TT_PointerOrReference;
746 
747     if (PrevToken->Tok.isLiteral() ||
748         PrevToken->isOneOf(tok::r_paren, tok::r_square) ||
749         NextToken->Tok.isLiteral() || NextToken->isUnaryOperator())
750       return TT_BinaryOperator;
751 
752     // It is very unlikely that we are going to find a pointer or reference type
753     // definition on the RHS of an assignment.
754     if (IsExpression)
755       return TT_BinaryOperator;
756 
757     return TT_PointerOrReference;
758   }
759 
determinePlusMinusCaretUsage(const FormatToken & Tok)760   TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
761     const FormatToken *PrevToken = Tok.getPreviousNonComment();
762     if (PrevToken == NULL || PrevToken->Type == TT_CastRParen)
763       return TT_UnaryOperator;
764 
765     // Use heuristics to recognize unary operators.
766     if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
767                            tok::question, tok::colon, tok::kw_return,
768                            tok::kw_case, tok::at, tok::l_brace))
769       return TT_UnaryOperator;
770 
771     // There can't be two consecutive binary operators.
772     if (PrevToken->Type == TT_BinaryOperator)
773       return TT_UnaryOperator;
774 
775     // Fall back to marking the token as binary operator.
776     return TT_BinaryOperator;
777   }
778 
779   /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
determineIncrementUsage(const FormatToken & Tok)780   TokenType determineIncrementUsage(const FormatToken &Tok) {
781     const FormatToken *PrevToken = Tok.getPreviousNonComment();
782     if (PrevToken == NULL || PrevToken->Type == TT_CastRParen)
783       return TT_UnaryOperator;
784     if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
785       return TT_TrailingUnaryOperator;
786 
787     return TT_UnaryOperator;
788   }
789 
790   // FIXME: This is copy&pasted from Sema. Put it in a common place and remove
791   // duplication.
792   /// \brief Determine whether the token kind starts a simple-type-specifier.
isSimpleTypeSpecifier(const FormatToken & Tok) const793   bool isSimpleTypeSpecifier(const FormatToken &Tok) const {
794     switch (Tok.Tok.getKind()) {
795     case tok::kw_short:
796     case tok::kw_long:
797     case tok::kw___int64:
798     case tok::kw___int128:
799     case tok::kw_signed:
800     case tok::kw_unsigned:
801     case tok::kw_void:
802     case tok::kw_char:
803     case tok::kw_int:
804     case tok::kw_half:
805     case tok::kw_float:
806     case tok::kw_double:
807     case tok::kw_wchar_t:
808     case tok::kw_bool:
809     case tok::kw___underlying_type:
810     case tok::annot_typename:
811     case tok::kw_char16_t:
812     case tok::kw_char32_t:
813     case tok::kw_typeof:
814     case tok::kw_decltype:
815       return true;
816     default:
817       return false;
818     }
819   }
820 
821   SmallVector<Context, 8> Contexts;
822 
823   AnnotatedLine &Line;
824   FormatToken *CurrentToken;
825   bool KeywordVirtualFound;
826   bool NameFound;
827   bool AutoFound;
828   IdentifierInfo &Ident_in;
829 };
830 
831 /// \brief Parses binary expressions by inserting fake parenthesis based on
832 /// operator precedence.
833 class ExpressionParser {
834 public:
ExpressionParser(AnnotatedLine & Line)835   ExpressionParser(AnnotatedLine &Line) : Current(Line.First) {
836     // Skip leading "}", e.g. in "} else if (...) {".
837     if (Current->is(tok::r_brace))
838       next();
839   }
840 
841   /// \brief Parse expressions with the given operatore precedence.
parse(int Precedence=0)842   void parse(int Precedence = 0) {
843     // Conditional expressions need to be parsed separately for proper nesting.
844     if (Precedence == prec::Conditional + 1) {
845       parseConditionalExpr();
846       return;
847     }
848     if (Precedence > prec::PointerToMember || Current == NULL)
849       return;
850 
851     FormatToken *Start = Current;
852     bool OperatorFound = false;
853 
854     while (Current) {
855       // Consume operators with higher precedence.
856       parse(Precedence + 1);
857 
858       int CurrentPrecedence = 0;
859       if (Current) {
860         if (Current->Type == TT_ConditionalExpr)
861           CurrentPrecedence = 1 + (int)prec::Conditional;
862         else if (Current->is(tok::semi) || Current->Type == TT_InlineASMColon)
863           CurrentPrecedence = 1;
864         else if (Current->Type == TT_BinaryOperator || Current->is(tok::comma))
865           CurrentPrecedence = 1 + (int)Current->getPrecedence();
866         else if (Current->Type == TT_ObjCSelectorName) {
867           CurrentPrecedence = 1 + (int)prec::Assignment;
868           if (Precedence == CurrentPrecedence)
869             Start = Current;
870         }
871       }
872 
873       // At the end of the line or when an operator with higher precedence is
874       // found, insert fake parenthesis and return.
875       if (Current == NULL || Current->closesScope() ||
876           (CurrentPrecedence != 0 && CurrentPrecedence < Precedence)) {
877         if (OperatorFound)
878           addFakeParenthesis(Start, prec::Level(Precedence - 1));
879         return;
880       }
881 
882       // Consume scopes: (), [], <> and {}
883       if (Current->opensScope()) {
884         while (Current && !Current->closesScope()) {
885           next();
886           parse();
887         }
888         next();
889       } else {
890         // Operator found.
891         if (CurrentPrecedence == Precedence)
892           OperatorFound = true;
893 
894         next();
895       }
896     }
897   }
898 
899 private:
addFakeParenthesis(FormatToken * Start,prec::Level Precedence)900   void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
901     Start->FakeLParens.push_back(Precedence);
902     if (Current)
903       ++Current->Previous->FakeRParens;
904   }
905 
parseConditionalExpr()906   void parseConditionalExpr() {
907     FormatToken *Start = Current;
908     parse(prec::LogicalOr + 1);
909     if (!Current || !Current->is(tok::question))
910       return;
911     next();
912     parse(prec::LogicalOr + 1);
913     if (!Current || Current->Type != TT_ConditionalExpr)
914       return;
915     next();
916     parseConditionalExpr();
917     addFakeParenthesis(Start, prec::Conditional);
918   }
919 
next()920   void next() {
921     if (Current)
922       Current = Current->Next;
923     while (Current && Current->isTrailingComment())
924       Current = Current->Next;
925   }
926 
927   FormatToken *Current;
928 };
929 
930 } // end anonymous namespace
931 
annotate(AnnotatedLine & Line)932 void TokenAnnotator::annotate(AnnotatedLine &Line) {
933   AnnotatingParser Parser(Line, Ident_in);
934   Line.Type = Parser.parseLine();
935   if (Line.Type == LT_Invalid)
936     return;
937 
938   ExpressionParser ExprParser(Line);
939   ExprParser.parse();
940 
941   if (Line.First->Type == TT_ObjCMethodSpecifier)
942     Line.Type = LT_ObjCMethodDecl;
943   else if (Line.First->Type == TT_ObjCDecl)
944     Line.Type = LT_ObjCDecl;
945   else if (Line.First->Type == TT_ObjCProperty)
946     Line.Type = LT_ObjCProperty;
947 
948   Line.First->SpacesRequiredBefore = 1;
949   Line.First->CanBreakBefore = Line.First->MustBreakBefore;
950 }
951 
calculateFormattingInformation(AnnotatedLine & Line)952 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
953   Line.First->TotalLength = Line.First->CodePointCount;
954   if (!Line.First->Next)
955     return;
956   FormatToken *Current = Line.First->Next;
957   while (Current != NULL) {
958     if (Current->Type == TT_LineComment)
959       Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
960     else
961       Current->SpacesRequiredBefore =
962           spaceRequiredBefore(Line, *Current) ? 1 : 0;
963 
964     if (Current->is(tok::comment)) {
965       Current->MustBreakBefore = Current->NewlinesBefore > 0;
966     } else if (Current->Previous->isTrailingComment() ||
967                (Current->is(tok::string_literal) &&
968                 Current->Previous->is(tok::string_literal))) {
969       Current->MustBreakBefore = true;
970     } else if (Current->Previous->IsUnterminatedLiteral) {
971       Current->MustBreakBefore = true;
972     } else if (Current->is(tok::lessless) && Current->Next &&
973                Current->Previous->is(tok::string_literal) &&
974                Current->Next->is(tok::string_literal)) {
975       Current->MustBreakBefore = true;
976     } else if (Current->Previous->ClosesTemplateDeclaration &&
977                Style.AlwaysBreakTemplateDeclarations) {
978       Current->MustBreakBefore = true;
979     } else if (Current->Type == TT_CtorInitializerComma &&
980                Style.BreakConstructorInitializersBeforeComma) {
981       Current->MustBreakBefore = true;
982     }
983     Current->CanBreakBefore =
984         Current->MustBreakBefore || canBreakBefore(Line, *Current);
985     if (Current->MustBreakBefore ||
986         (Current->is(tok::string_literal) &&
987          Current->TokenText.find("\\\n") != StringRef::npos))
988       Current->TotalLength = Current->Previous->TotalLength + Style.ColumnLimit;
989     else
990       Current->TotalLength = Current->Previous->TotalLength +
991                              Current->CodePointCount +
992                              Current->SpacesRequiredBefore;
993     // FIXME: Only calculate this if CanBreakBefore is true once static
994     // initializers etc. are sorted out.
995     // FIXME: Move magic numbers to a better place.
996     Current->SplitPenalty =
997         20 * Current->BindingStrength + splitPenalty(Line, *Current);
998 
999     Current = Current->Next;
1000   }
1001 
1002   calculateUnbreakableTailLengths(Line);
1003   DEBUG({
1004     printDebugInfo(Line);
1005   });
1006 }
1007 
calculateUnbreakableTailLengths(AnnotatedLine & Line)1008 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1009   unsigned UnbreakableTailLength = 0;
1010   FormatToken *Current = Line.Last;
1011   while (Current != NULL) {
1012     Current->UnbreakableTailLength = UnbreakableTailLength;
1013     if (Current->CanBreakBefore ||
1014         Current->isOneOf(tok::comment, tok::string_literal)) {
1015       UnbreakableTailLength = 0;
1016     } else {
1017       UnbreakableTailLength +=
1018           Current->CodePointCount + Current->SpacesRequiredBefore;
1019     }
1020     Current = Current->Previous;
1021   }
1022 }
1023 
splitPenalty(const AnnotatedLine & Line,const FormatToken & Tok)1024 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1025                                       const FormatToken &Tok) {
1026   const FormatToken &Left = *Tok.Previous;
1027   const FormatToken &Right = Tok;
1028 
1029   if (Left.is(tok::semi))
1030     return 0;
1031   if (Left.is(tok::comma))
1032     return 1;
1033   if (Right.is(tok::l_square))
1034     return 150;
1035 
1036   if (Right.Type == TT_StartOfName || Right.is(tok::kw_operator)) {
1037     if (Line.First->is(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1038       return 3;
1039     if (Left.Type == TT_StartOfName)
1040       return 20;
1041     else if (Line.MightBeFunctionDecl && Right.BindingStrength == 1)
1042       // FIXME: Clean up hack of using BindingStrength to find top-level names.
1043       return Style.PenaltyReturnTypeOnItsOwnLine;
1044     else
1045       return 200;
1046   }
1047   if (Left.is(tok::equal) && Right.is(tok::l_brace))
1048     return 150;
1049   if (Left.Type == TT_CastRParen)
1050     return 100;
1051   if (Left.is(tok::coloncolon))
1052     return 500;
1053   if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1054     return 5000;
1055 
1056   if (Left.Type == TT_RangeBasedForLoopColon ||
1057       Left.Type == TT_InheritanceColon)
1058     return 2;
1059 
1060   if (Right.isOneOf(tok::arrow, tok::period) &&
1061       Right.Type != TT_DesignatedInitializerPeriod) {
1062     if (Left.isOneOf(tok::r_paren, tok::r_square) && Left.MatchingParen &&
1063         Left.MatchingParen->ParameterCount > 0)
1064       return 20; // Should be smaller than breaking at a nested comma.
1065     return 150;
1066   }
1067 
1068   // Breaking before a trailing 'const' or not-function-like annotation is bad.
1069   if (Left.is(tok::r_paren) && Line.Type != LT_ObjCProperty &&
1070       (Right.is(tok::kw_const) || (Right.is(tok::identifier) && Right.Next &&
1071                                    Right.Next->isNot(tok::l_paren))))
1072     return 150;
1073 
1074   // In for-loops, prefer breaking at ',' and ';'.
1075   if (Line.First->is(tok::kw_for) && Left.is(tok::equal))
1076     return 4;
1077 
1078   // In Objective-C method expressions, prefer breaking before "param:" over
1079   // breaking after it.
1080   if (Right.Type == TT_ObjCSelectorName)
1081     return 0;
1082   if (Left.is(tok::colon) && Left.Type == TT_ObjCMethodExpr)
1083     return 20;
1084 
1085   if (Left.is(tok::l_paren) && Line.MightBeFunctionDecl)
1086     return 100;
1087   if (Left.opensScope())
1088     return Left.ParameterCount > 1 ? prec::Comma : 20;
1089 
1090   if (Right.is(tok::lessless)) {
1091     if (Left.is(tok::string_literal)) {
1092       StringRef Content = Left.TokenText;
1093       Content = Content.drop_back(1).drop_front(1).trim();
1094       if (Content.size() > 1 &&
1095           (Content.back() == ':' || Content.back() == '='))
1096         return 25;
1097     }
1098     return 1; // Breaking at a << is really cheap.
1099   }
1100   if (Left.Type == TT_ConditionalExpr)
1101     return prec::Conditional;
1102   prec::Level Level = Left.getPrecedence();
1103 
1104   if (Level != prec::Unknown)
1105     return Level;
1106 
1107   return 3;
1108 }
1109 
spaceRequiredBetween(const AnnotatedLine & Line,const FormatToken & Left,const FormatToken & Right)1110 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
1111                                           const FormatToken &Left,
1112                                           const FormatToken &Right) {
1113   if (Right.is(tok::hashhash))
1114     return Left.is(tok::hash);
1115   if (Left.isOneOf(tok::hashhash, tok::hash))
1116     return Right.is(tok::hash);
1117   if (Right.isOneOf(tok::r_paren, tok::semi, tok::comma))
1118     return false;
1119   if (Right.is(tok::less) &&
1120       (Left.is(tok::kw_template) ||
1121        (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)))
1122     return true;
1123   if (Left.is(tok::arrow) || Right.is(tok::arrow))
1124     return false;
1125   if (Left.isOneOf(tok::exclaim, tok::tilde))
1126     return false;
1127   if (Left.is(tok::at) &&
1128       Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
1129                     tok::numeric_constant, tok::l_paren, tok::l_brace,
1130                     tok::kw_true, tok::kw_false))
1131     return false;
1132   if (Left.is(tok::coloncolon))
1133     return false;
1134   if (Right.is(tok::coloncolon))
1135     return !Left.isOneOf(tok::identifier, tok::greater, tok::l_paren);
1136   if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
1137     return false;
1138   if (Right.is(tok::ellipsis))
1139     return false;
1140   if (Right.Type == TT_PointerOrReference)
1141     return Left.Tok.isLiteral() ||
1142            ((Left.Type != TT_PointerOrReference) && Left.isNot(tok::l_paren) &&
1143             !Style.PointerBindsToType);
1144   if (Right.Type == TT_FunctionTypeLParen && Left.isNot(tok::l_paren) &&
1145       (Left.Type != TT_PointerOrReference || Style.PointerBindsToType))
1146     return true;
1147   if (Left.Type == TT_PointerOrReference)
1148     return Right.Tok.isLiteral() || Right.Type == TT_BlockComment ||
1149            ((Right.Type != TT_PointerOrReference) &&
1150             Right.isNot(tok::l_paren) && Style.PointerBindsToType &&
1151             Left.Previous &&
1152             !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
1153   if (Right.is(tok::star) && Left.is(tok::l_paren))
1154     return false;
1155   if (Left.is(tok::l_square))
1156     return Left.Type == TT_ObjCArrayLiteral && Right.isNot(tok::r_square);
1157   if (Right.is(tok::r_square))
1158     return Right.Type == TT_ObjCArrayLiteral;
1159   if (Right.is(tok::l_square) && Right.Type != TT_ObjCMethodExpr)
1160     return false;
1161   if (Left.is(tok::colon))
1162     return Left.Type != TT_ObjCMethodExpr;
1163   if (Right.is(tok::colon))
1164     return Right.Type != TT_ObjCMethodExpr && !Left.is(tok::question);
1165   if (Left.is(tok::l_paren))
1166     return false;
1167   if (Right.is(tok::l_paren)) {
1168     if (Left.is(tok::r_paren) && Left.MatchingParen &&
1169         Left.MatchingParen->Previous &&
1170         Left.MatchingParen->Previous->is(tok::kw___attribute))
1171       return true;
1172     return Line.Type == LT_ObjCDecl ||
1173            Left.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch,
1174                         tok::kw_return, tok::kw_catch, tok::kw_new,
1175                         tok::kw_delete, tok::semi);
1176   }
1177   if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
1178     return false;
1179   if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
1180     return false; // No spaces in "{}".
1181   if (Left.is(tok::l_brace) || Right.is(tok::r_brace))
1182     return !Style.Cpp11BracedListStyle;
1183   if (Right.Type == TT_UnaryOperator)
1184     return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
1185            (Left.isNot(tok::colon) || Left.Type != TT_ObjCMethodExpr);
1186   if (Left.isOneOf(tok::identifier, tok::greater, tok::r_square) &&
1187       Right.is(tok::l_brace) && Right.getNextNonComment())
1188     return false;
1189   if (Left.is(tok::period) || Right.is(tok::period))
1190     return false;
1191   if (Left.Type == TT_BlockComment && Left.TokenText.endswith("=*/"))
1192     return false;
1193   return true;
1194 }
1195 
spaceRequiredBefore(const AnnotatedLine & Line,const FormatToken & Tok)1196 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
1197                                          const FormatToken &Tok) {
1198   if (Tok.Tok.getIdentifierInfo() && Tok.Previous->Tok.getIdentifierInfo())
1199     return true; // Never ever merge two identifiers.
1200   if (Line.Type == LT_ObjCMethodDecl) {
1201     if (Tok.Previous->Type == TT_ObjCMethodSpecifier)
1202       return true;
1203     if (Tok.Previous->is(tok::r_paren) && Tok.is(tok::identifier))
1204       // Don't space between ')' and <id>
1205       return false;
1206   }
1207   if (Line.Type == LT_ObjCProperty &&
1208       (Tok.is(tok::equal) || Tok.Previous->is(tok::equal)))
1209     return false;
1210 
1211   if (Tok.Type == TT_TrailingReturnArrow ||
1212       Tok.Previous->Type == TT_TrailingReturnArrow)
1213     return true;
1214   if (Tok.Previous->is(tok::comma))
1215     return true;
1216   if (Tok.is(tok::comma))
1217     return false;
1218   if (Tok.Type == TT_CtorInitializerColon || Tok.Type == TT_ObjCBlockLParen)
1219     return true;
1220   if (Tok.Previous->Tok.is(tok::kw_operator))
1221     return false;
1222   if (Tok.Type == TT_OverloadedOperatorLParen)
1223     return false;
1224   if (Tok.is(tok::colon))
1225     return !Line.First->isOneOf(tok::kw_case, tok::kw_default) &&
1226            Tok.getNextNonComment() != NULL && Tok.Type != TT_ObjCMethodExpr &&
1227            !Tok.Previous->is(tok::question);
1228   if (Tok.Previous->Type == TT_UnaryOperator ||
1229       Tok.Previous->Type == TT_CastRParen)
1230     return false;
1231   if (Tok.Previous->is(tok::greater) && Tok.is(tok::greater)) {
1232     return Tok.Type == TT_TemplateCloser &&
1233            Tok.Previous->Type == TT_TemplateCloser &&
1234            Style.Standard != FormatStyle::LS_Cpp11;
1235   }
1236   if (Tok.isOneOf(tok::arrowstar, tok::periodstar) ||
1237       Tok.Previous->isOneOf(tok::arrowstar, tok::periodstar))
1238     return false;
1239   if ((Tok.Type == TT_BinaryOperator && !Tok.Previous->is(tok::l_paren)) ||
1240       Tok.Previous->Type == TT_BinaryOperator)
1241     return true;
1242   if (Tok.Previous->Type == TT_TemplateCloser && Tok.is(tok::l_paren))
1243     return false;
1244   if (Tok.is(tok::less) && Line.First->is(tok::hash))
1245     return true;
1246   if (Tok.Type == TT_TrailingUnaryOperator)
1247     return false;
1248   return spaceRequiredBetween(Line, *Tok.Previous, Tok);
1249 }
1250 
canBreakBefore(const AnnotatedLine & Line,const FormatToken & Right)1251 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
1252                                     const FormatToken &Right) {
1253   const FormatToken &Left = *Right.Previous;
1254   if (Right.Type == TT_StartOfName || Right.is(tok::kw_operator))
1255     return true;
1256   if (Right.is(tok::colon) &&
1257       (Right.Type == TT_ObjCDictLiteral || Right.Type == TT_ObjCMethodExpr))
1258     return false;
1259   if (Left.is(tok::colon) &&
1260       (Left.Type == TT_ObjCDictLiteral || Left.Type == TT_ObjCMethodExpr))
1261     return true;
1262   if (Right.Type == TT_ObjCSelectorName)
1263     return true;
1264   if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
1265     return true;
1266   if (Left.ClosesTemplateDeclaration)
1267     return true;
1268   if ((Right.Type == TT_ConditionalExpr &&
1269        !(Right.is(tok::colon) && Left.is(tok::question))) ||
1270       Right.is(tok::question))
1271     return true;
1272   if (Right.Type == TT_RangeBasedForLoopColon ||
1273       Right.Type == TT_OverloadedOperatorLParen)
1274     return false;
1275   if (Left.Type == TT_RangeBasedForLoopColon)
1276     return true;
1277   if (Right.Type == TT_RangeBasedForLoopColon)
1278     return false;
1279   if (Left.Type == TT_PointerOrReference || Left.Type == TT_TemplateCloser ||
1280       Left.Type == TT_UnaryOperator || Left.Type == TT_ConditionalExpr ||
1281       Left.isOneOf(tok::question, tok::kw_operator))
1282     return false;
1283   if (Left.is(tok::equal) && Line.Type == LT_VirtualFunctionDecl)
1284     return false;
1285   if (Left.Previous) {
1286     if (Left.is(tok::l_paren) && Right.is(tok::l_paren) &&
1287         Left.Previous->is(tok::kw___attribute))
1288       return false;
1289     if (Left.is(tok::l_paren) && (Left.Previous->Type == TT_BinaryOperator ||
1290                                   Left.Previous->Type == TT_CastRParen))
1291       return false;
1292   }
1293 
1294   if (Right.isTrailingComment())
1295     // We rely on MustBreakBefore being set correctly here as we should not
1296     // change the "binding" behavior of a comment.
1297     return false;
1298 
1299   // We only break before r_brace if there was a corresponding break before
1300   // the l_brace, which is tracked by BreakBeforeClosingBrace.
1301   if (Right.isOneOf(tok::r_brace, tok::r_paren) ||
1302       Right.Type == TT_TemplateCloser)
1303     return false;
1304 
1305   // Allow breaking after a trailing 'const', e.g. after a method declaration,
1306   // unless it is follow by ';', '{' or '='.
1307   if (Left.is(tok::kw_const) && Left.Previous != NULL &&
1308       Left.Previous->is(tok::r_paren))
1309     return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal);
1310 
1311   if (Right.is(tok::kw___attribute))
1312     return true;
1313 
1314   if (Left.is(tok::identifier) && Right.is(tok::string_literal))
1315     return true;
1316 
1317   if (Left.Type == TT_CtorInitializerComma &&
1318       Style.BreakConstructorInitializersBeforeComma)
1319     return false;
1320   if (Right.isBinaryOperator() && Style.BreakBeforeBinaryOperators)
1321     return true;
1322   return (Left.isBinaryOperator() && Left.isNot(tok::lessless) &&
1323           !Style.BreakBeforeBinaryOperators) ||
1324          Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
1325                       tok::kw_class, tok::kw_struct) ||
1326          Right.isOneOf(tok::lessless, tok::arrow, tok::period, tok::colon) ||
1327          (Left.is(tok::r_paren) &&
1328           Right.isOneOf(tok::identifier, tok::kw_const, tok::kw___attribute)) ||
1329          (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) ||
1330          Right.is(tok::l_square);
1331 }
1332 
printDebugInfo(const AnnotatedLine & Line)1333 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
1334   llvm::errs() << "AnnotatedTokens:\n";
1335   const FormatToken *Tok = Line.First;
1336   while (Tok) {
1337     llvm::errs() << " M=" << Tok->MustBreakBefore
1338                  << " C=" << Tok->CanBreakBefore << " T=" << Tok->Type
1339                  << " S=" << Tok->SpacesRequiredBefore
1340                  << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
1341                  << " PPK=" << Tok->PackingKind << " FakeLParens=";
1342     for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
1343       llvm::errs() << Tok->FakeLParens[i] << "/";
1344     llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n";
1345     Tok = Tok->Next;
1346   }
1347   llvm::errs() << "----\n";
1348 }
1349 
1350 } // namespace format
1351 } // namespace clang
1352