xref: /freebsd/contrib/llvm-project/clang/lib/Format/UnwrappedLineParser.cpp (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file contains the implementation of the UnwrappedLineParser,
11 /// which turns a stream of tokens into UnwrappedLines.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "UnwrappedLineParser.h"
16 #include "FormatToken.h"
17 #include "TokenAnnotator.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/Support/Debug.h"
20 #include "llvm/Support/raw_ostream.h"
21 
22 #include <algorithm>
23 
24 #define DEBUG_TYPE "format-parser"
25 
26 namespace clang {
27 namespace format {
28 
29 class FormatTokenSource {
30 public:
31   virtual ~FormatTokenSource() {}
32 
33   // Returns the next token in the token stream.
34   virtual FormatToken *getNextToken() = 0;
35 
36   // Returns the token preceding the token returned by the last call to
37   // getNextToken() in the token stream, or nullptr if no such token exists.
38   virtual FormatToken *getPreviousToken() = 0;
39 
40   // Returns the token that would be returned by the next call to
41   // getNextToken().
42   virtual FormatToken *peekNextToken() = 0;
43 
44   // Returns whether we are at the end of the file.
45   // This can be different from whether getNextToken() returned an eof token
46   // when the FormatTokenSource is a view on a part of the token stream.
47   virtual bool isEOF() = 0;
48 
49   // Gets the current position in the token stream, to be used by setPosition().
50   virtual unsigned getPosition() = 0;
51 
52   // Resets the token stream to the state it was in when getPosition() returned
53   // Position, and return the token at that position in the stream.
54   virtual FormatToken *setPosition(unsigned Position) = 0;
55 };
56 
57 namespace {
58 
59 class ScopedDeclarationState {
60 public:
61   ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack,
62                          bool MustBeDeclaration)
63       : Line(Line), Stack(Stack) {
64     Line.MustBeDeclaration = MustBeDeclaration;
65     Stack.push_back(MustBeDeclaration);
66   }
67   ~ScopedDeclarationState() {
68     Stack.pop_back();
69     if (!Stack.empty())
70       Line.MustBeDeclaration = Stack.back();
71     else
72       Line.MustBeDeclaration = true;
73   }
74 
75 private:
76   UnwrappedLine &Line;
77   llvm::BitVector &Stack;
78 };
79 
80 static bool isLineComment(const FormatToken &FormatTok) {
81   return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*");
82 }
83 
84 // Checks if \p FormatTok is a line comment that continues the line comment
85 // \p Previous. The original column of \p MinColumnToken is used to determine
86 // whether \p FormatTok is indented enough to the right to continue \p Previous.
87 static bool continuesLineComment(const FormatToken &FormatTok,
88                                  const FormatToken *Previous,
89                                  const FormatToken *MinColumnToken) {
90   if (!Previous || !MinColumnToken)
91     return false;
92   unsigned MinContinueColumn =
93       MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1);
94   return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 &&
95          isLineComment(*Previous) &&
96          FormatTok.OriginalColumn >= MinContinueColumn;
97 }
98 
99 class ScopedMacroState : public FormatTokenSource {
100 public:
101   ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource,
102                    FormatToken *&ResetToken)
103       : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken),
104         PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource),
105         Token(nullptr), PreviousToken(nullptr) {
106     FakeEOF.Tok.startToken();
107     FakeEOF.Tok.setKind(tok::eof);
108     TokenSource = this;
109     Line.Level = 0;
110     Line.InPPDirective = true;
111   }
112 
113   ~ScopedMacroState() override {
114     TokenSource = PreviousTokenSource;
115     ResetToken = Token;
116     Line.InPPDirective = false;
117     Line.Level = PreviousLineLevel;
118   }
119 
120   FormatToken *getNextToken() override {
121     // The \c UnwrappedLineParser guards against this by never calling
122     // \c getNextToken() after it has encountered the first eof token.
123     assert(!eof());
124     PreviousToken = Token;
125     Token = PreviousTokenSource->getNextToken();
126     if (eof())
127       return &FakeEOF;
128     return Token;
129   }
130 
131   FormatToken *getPreviousToken() override {
132     return PreviousTokenSource->getPreviousToken();
133   }
134 
135   FormatToken *peekNextToken() override {
136     if (eof())
137       return &FakeEOF;
138     return PreviousTokenSource->peekNextToken();
139   }
140 
141   bool isEOF() override { return PreviousTokenSource->isEOF(); }
142 
143   unsigned getPosition() override { return PreviousTokenSource->getPosition(); }
144 
145   FormatToken *setPosition(unsigned Position) override {
146     PreviousToken = nullptr;
147     Token = PreviousTokenSource->setPosition(Position);
148     return Token;
149   }
150 
151 private:
152   bool eof() {
153     return Token && Token->HasUnescapedNewline &&
154            !continuesLineComment(*Token, PreviousToken,
155                                  /*MinColumnToken=*/PreviousToken);
156   }
157 
158   FormatToken FakeEOF;
159   UnwrappedLine &Line;
160   FormatTokenSource *&TokenSource;
161   FormatToken *&ResetToken;
162   unsigned PreviousLineLevel;
163   FormatTokenSource *PreviousTokenSource;
164 
165   FormatToken *Token;
166   FormatToken *PreviousToken;
167 };
168 
169 } // end anonymous namespace
170 
171 class ScopedLineState {
172 public:
173   ScopedLineState(UnwrappedLineParser &Parser,
174                   bool SwitchToPreprocessorLines = false)
175       : Parser(Parser), OriginalLines(Parser.CurrentLines) {
176     if (SwitchToPreprocessorLines)
177       Parser.CurrentLines = &Parser.PreprocessorDirectives;
178     else if (!Parser.Line->Tokens.empty())
179       Parser.CurrentLines = &Parser.Line->Tokens.back().Children;
180     PreBlockLine = std::move(Parser.Line);
181     Parser.Line = std::make_unique<UnwrappedLine>();
182     Parser.Line->Level = PreBlockLine->Level;
183     Parser.Line->InPPDirective = PreBlockLine->InPPDirective;
184   }
185 
186   ~ScopedLineState() {
187     if (!Parser.Line->Tokens.empty()) {
188       Parser.addUnwrappedLine();
189     }
190     assert(Parser.Line->Tokens.empty());
191     Parser.Line = std::move(PreBlockLine);
192     if (Parser.CurrentLines == &Parser.PreprocessorDirectives)
193       Parser.MustBreakBeforeNextToken = true;
194     Parser.CurrentLines = OriginalLines;
195   }
196 
197 private:
198   UnwrappedLineParser &Parser;
199 
200   std::unique_ptr<UnwrappedLine> PreBlockLine;
201   SmallVectorImpl<UnwrappedLine> *OriginalLines;
202 };
203 
204 class CompoundStatementIndenter {
205 public:
206   CompoundStatementIndenter(UnwrappedLineParser *Parser,
207                             const FormatStyle &Style, unsigned &LineLevel)
208       : CompoundStatementIndenter(Parser, LineLevel,
209                                   Style.BraceWrapping.AfterControlStatement,
210                                   Style.BraceWrapping.IndentBraces) {}
211   CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel,
212                             bool WrapBrace, bool IndentBrace)
213       : LineLevel(LineLevel), OldLineLevel(LineLevel) {
214     if (WrapBrace)
215       Parser->addUnwrappedLine();
216     if (IndentBrace)
217       ++LineLevel;
218   }
219   ~CompoundStatementIndenter() { LineLevel = OldLineLevel; }
220 
221 private:
222   unsigned &LineLevel;
223   unsigned OldLineLevel;
224 };
225 
226 namespace {
227 
228 class IndexedTokenSource : public FormatTokenSource {
229 public:
230   IndexedTokenSource(ArrayRef<FormatToken *> Tokens)
231       : Tokens(Tokens), Position(-1) {}
232 
233   FormatToken *getNextToken() override {
234     if (Position >= 0 && Tokens[Position]->is(tok::eof)) {
235       LLVM_DEBUG({
236         llvm::dbgs() << "Next ";
237         dbgToken(Position);
238       });
239       return Tokens[Position];
240     }
241     ++Position;
242     LLVM_DEBUG({
243       llvm::dbgs() << "Next ";
244       dbgToken(Position);
245     });
246     return Tokens[Position];
247   }
248 
249   FormatToken *getPreviousToken() override {
250     return Position > 0 ? Tokens[Position - 1] : nullptr;
251   }
252 
253   FormatToken *peekNextToken() override {
254     int Next = Position + 1;
255     LLVM_DEBUG({
256       llvm::dbgs() << "Peeking ";
257       dbgToken(Next);
258     });
259     return Tokens[Next];
260   }
261 
262   bool isEOF() override { return Tokens[Position]->is(tok::eof); }
263 
264   unsigned getPosition() override {
265     LLVM_DEBUG(llvm::dbgs() << "Getting Position: " << Position << "\n");
266     assert(Position >= 0);
267     return Position;
268   }
269 
270   FormatToken *setPosition(unsigned P) override {
271     LLVM_DEBUG(llvm::dbgs() << "Setting Position: " << P << "\n");
272     Position = P;
273     return Tokens[Position];
274   }
275 
276   void reset() { Position = -1; }
277 
278 private:
279   void dbgToken(int Position, llvm::StringRef Indent = "") {
280     FormatToken *Tok = Tokens[Position];
281     llvm::dbgs() << Indent << "[" << Position
282                  << "] Token: " << Tok->Tok.getName() << " / " << Tok->TokenText
283                  << ", Macro: " << !!Tok->MacroCtx << "\n";
284   }
285 
286   ArrayRef<FormatToken *> Tokens;
287   int Position;
288 };
289 
290 } // end anonymous namespace
291 
292 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style,
293                                          const AdditionalKeywords &Keywords,
294                                          unsigned FirstStartColumn,
295                                          ArrayRef<FormatToken *> Tokens,
296                                          UnwrappedLineConsumer &Callback)
297     : Line(new UnwrappedLine), MustBreakBeforeNextToken(false),
298       CurrentLines(&Lines), Style(Style), Keywords(Keywords),
299       CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr),
300       Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1),
301       IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None
302                        ? IG_Rejected
303                        : IG_Inited),
304       IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {}
305 
306 void UnwrappedLineParser::reset() {
307   PPBranchLevel = -1;
308   IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None
309                      ? IG_Rejected
310                      : IG_Inited;
311   IncludeGuardToken = nullptr;
312   Line.reset(new UnwrappedLine);
313   CommentsBeforeNextToken.clear();
314   FormatTok = nullptr;
315   MustBreakBeforeNextToken = false;
316   PreprocessorDirectives.clear();
317   CurrentLines = &Lines;
318   DeclarationScopeStack.clear();
319   NestedTooDeep.clear();
320   PPStack.clear();
321   Line->FirstStartColumn = FirstStartColumn;
322 }
323 
324 void UnwrappedLineParser::parse() {
325   IndexedTokenSource TokenSource(AllTokens);
326   Line->FirstStartColumn = FirstStartColumn;
327   do {
328     LLVM_DEBUG(llvm::dbgs() << "----\n");
329     reset();
330     Tokens = &TokenSource;
331     TokenSource.reset();
332 
333     readToken();
334     parseFile();
335 
336     // If we found an include guard then all preprocessor directives (other than
337     // the guard) are over-indented by one.
338     if (IncludeGuard == IG_Found)
339       for (auto &Line : Lines)
340         if (Line.InPPDirective && Line.Level > 0)
341           --Line.Level;
342 
343     // Create line with eof token.
344     pushToken(FormatTok);
345     addUnwrappedLine();
346 
347     for (const UnwrappedLine &Line : Lines)
348       Callback.consumeUnwrappedLine(Line);
349 
350     Callback.finishRun();
351     Lines.clear();
352     while (!PPLevelBranchIndex.empty() &&
353            PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) {
354       PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1);
355       PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1);
356     }
357     if (!PPLevelBranchIndex.empty()) {
358       ++PPLevelBranchIndex.back();
359       assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size());
360       assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back());
361     }
362   } while (!PPLevelBranchIndex.empty());
363 }
364 
365 void UnwrappedLineParser::parseFile() {
366   // The top-level context in a file always has declarations, except for pre-
367   // processor directives and JavaScript files.
368   bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript();
369   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
370                                           MustBeDeclaration);
371   if (Style.Language == FormatStyle::LK_TextProto)
372     parseBracedList();
373   else
374     parseLevel(/*HasOpeningBrace=*/false);
375   // Make sure to format the remaining tokens.
376   //
377   // LK_TextProto is special since its top-level is parsed as the body of a
378   // braced list, which does not necessarily have natural line separators such
379   // as a semicolon. Comments after the last entry that have been determined to
380   // not belong to that line, as in:
381   //   key: value
382   //   // endfile comment
383   // do not have a chance to be put on a line of their own until this point.
384   // Here we add this newline before end-of-file comments.
385   if (Style.Language == FormatStyle::LK_TextProto &&
386       !CommentsBeforeNextToken.empty())
387     addUnwrappedLine();
388   flushComments(true);
389   addUnwrappedLine();
390 }
391 
392 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() {
393   do {
394     switch (FormatTok->Tok.getKind()) {
395     case tok::l_brace:
396       return;
397     default:
398       if (FormatTok->is(Keywords.kw_where)) {
399         addUnwrappedLine();
400         nextToken();
401         parseCSharpGenericTypeConstraint();
402         break;
403       }
404       nextToken();
405       break;
406     }
407   } while (!eof());
408 }
409 
410 void UnwrappedLineParser::parseCSharpAttribute() {
411   int UnpairedSquareBrackets = 1;
412   do {
413     switch (FormatTok->Tok.getKind()) {
414     case tok::r_square:
415       nextToken();
416       --UnpairedSquareBrackets;
417       if (UnpairedSquareBrackets == 0) {
418         addUnwrappedLine();
419         return;
420       }
421       break;
422     case tok::l_square:
423       ++UnpairedSquareBrackets;
424       nextToken();
425       break;
426     default:
427       nextToken();
428       break;
429     }
430   } while (!eof());
431 }
432 
433 bool UnwrappedLineParser::precededByCommentOrPPDirective() const {
434   if (!Lines.empty() && Lines.back().InPPDirective)
435     return true;
436 
437   const FormatToken *Previous = Tokens->getPreviousToken();
438   return Previous && Previous->is(tok::comment) &&
439          (Previous->IsMultiline || Previous->NewlinesBefore > 0);
440 }
441 
442 bool UnwrappedLineParser::mightFitOnOneLine() const {
443   const auto ColumnLimit = Style.ColumnLimit;
444   if (ColumnLimit == 0)
445     return true;
446 
447   if (Lines.empty())
448     return true;
449 
450   const auto &PreviousLine = Lines.back();
451   const auto &Tokens = PreviousLine.Tokens;
452   assert(!Tokens.empty());
453   const auto *LastToken = Tokens.back().Tok;
454   assert(LastToken);
455   if (!LastToken->isOneOf(tok::semi, tok::comment))
456     return true;
457 
458   AnnotatedLine Line(PreviousLine);
459   assert(Line.Last == LastToken);
460 
461   TokenAnnotator Annotator(Style, Keywords);
462   Annotator.annotate(Line);
463   Annotator.calculateFormattingInformation(Line);
464 
465   return Line.Level * Style.IndentWidth + LastToken->TotalLength <= ColumnLimit;
466 }
467 
468 // Returns true if a simple block, or false otherwise. (A simple block has a
469 // single statement that fits on a single line.)
470 bool UnwrappedLineParser::parseLevel(bool HasOpeningBrace, IfStmtKind *IfKind) {
471   const bool IsPrecededByCommentOrPPDirective =
472       !Style.RemoveBracesLLVM || precededByCommentOrPPDirective();
473   unsigned StatementCount = 0;
474   bool SwitchLabelEncountered = false;
475   do {
476     tok::TokenKind kind = FormatTok->Tok.getKind();
477     if (FormatTok->getType() == TT_MacroBlockBegin) {
478       kind = tok::l_brace;
479     } else if (FormatTok->getType() == TT_MacroBlockEnd) {
480       kind = tok::r_brace;
481     }
482 
483     switch (kind) {
484     case tok::comment:
485       nextToken();
486       addUnwrappedLine();
487       break;
488     case tok::l_brace:
489       // FIXME: Add parameter whether this can happen - if this happens, we must
490       // be in a non-declaration context.
491       if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList())
492         continue;
493       parseBlock();
494       ++StatementCount;
495       assert(StatementCount > 0 && "StatementCount overflow!");
496       addUnwrappedLine();
497       break;
498     case tok::r_brace:
499       if (HasOpeningBrace) {
500         if (!Style.RemoveBracesLLVM)
501           return false;
502         if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 ||
503             IsPrecededByCommentOrPPDirective ||
504             precededByCommentOrPPDirective()) {
505           return false;
506         }
507         const FormatToken *Next = Tokens->peekNextToken();
508         if (Next->is(tok::comment) && Next->NewlinesBefore == 0)
509           return false;
510         return mightFitOnOneLine();
511       }
512       nextToken();
513       addUnwrappedLine();
514       break;
515     case tok::kw_default: {
516       unsigned StoredPosition = Tokens->getPosition();
517       FormatToken *Next;
518       do {
519         Next = Tokens->getNextToken();
520       } while (Next->is(tok::comment));
521       FormatTok = Tokens->setPosition(StoredPosition);
522       if (Next && Next->isNot(tok::colon)) {
523         // default not followed by ':' is not a case label; treat it like
524         // an identifier.
525         parseStructuralElement();
526         break;
527       }
528       // Else, if it is 'default:', fall through to the case handling.
529       LLVM_FALLTHROUGH;
530     }
531     case tok::kw_case:
532       if (Style.isJavaScript() && Line->MustBeDeclaration) {
533         // A 'case: string' style field declaration.
534         parseStructuralElement();
535         break;
536       }
537       if (!SwitchLabelEncountered &&
538           (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1)))
539         ++Line->Level;
540       SwitchLabelEncountered = true;
541       parseStructuralElement();
542       break;
543     case tok::l_square:
544       if (Style.isCSharp()) {
545         nextToken();
546         parseCSharpAttribute();
547         break;
548       }
549       LLVM_FALLTHROUGH;
550     default:
551       parseStructuralElement(IfKind, !HasOpeningBrace);
552       ++StatementCount;
553       assert(StatementCount > 0 && "StatementCount overflow!");
554       break;
555     }
556   } while (!eof());
557   return false;
558 }
559 
560 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
561   // We'll parse forward through the tokens until we hit
562   // a closing brace or eof - note that getNextToken() will
563   // parse macros, so this will magically work inside macro
564   // definitions, too.
565   unsigned StoredPosition = Tokens->getPosition();
566   FormatToken *Tok = FormatTok;
567   const FormatToken *PrevTok = Tok->Previous;
568   // Keep a stack of positions of lbrace tokens. We will
569   // update information about whether an lbrace starts a
570   // braced init list or a different block during the loop.
571   SmallVector<FormatToken *, 8> LBraceStack;
572   assert(Tok->Tok.is(tok::l_brace));
573   do {
574     // Get next non-comment token.
575     FormatToken *NextTok;
576     unsigned ReadTokens = 0;
577     do {
578       NextTok = Tokens->getNextToken();
579       ++ReadTokens;
580     } while (NextTok->is(tok::comment));
581 
582     switch (Tok->Tok.getKind()) {
583     case tok::l_brace:
584       if (Style.isJavaScript() && PrevTok) {
585         if (PrevTok->isOneOf(tok::colon, tok::less))
586           // A ':' indicates this code is in a type, or a braced list
587           // following a label in an object literal ({a: {b: 1}}).
588           // A '<' could be an object used in a comparison, but that is nonsense
589           // code (can never return true), so more likely it is a generic type
590           // argument (`X<{a: string; b: number}>`).
591           // The code below could be confused by semicolons between the
592           // individual members in a type member list, which would normally
593           // trigger BK_Block. In both cases, this must be parsed as an inline
594           // braced init.
595           Tok->setBlockKind(BK_BracedInit);
596         else if (PrevTok->is(tok::r_paren))
597           // `) { }` can only occur in function or method declarations in JS.
598           Tok->setBlockKind(BK_Block);
599       } else {
600         Tok->setBlockKind(BK_Unknown);
601       }
602       LBraceStack.push_back(Tok);
603       break;
604     case tok::r_brace:
605       if (LBraceStack.empty())
606         break;
607       if (LBraceStack.back()->is(BK_Unknown)) {
608         bool ProbablyBracedList = false;
609         if (Style.Language == FormatStyle::LK_Proto) {
610           ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square);
611         } else {
612           // Skip NextTok over preprocessor lines, otherwise we may not
613           // properly diagnose the block as a braced intializer
614           // if the comma separator appears after the pp directive.
615           while (NextTok->is(tok::hash)) {
616             ScopedMacroState MacroState(*Line, Tokens, NextTok);
617             do {
618               NextTok = Tokens->getNextToken();
619               ++ReadTokens;
620             } while (NextTok->isNot(tok::eof));
621           }
622 
623           // Using OriginalColumn to distinguish between ObjC methods and
624           // binary operators is a bit hacky.
625           bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) &&
626                                   NextTok->OriginalColumn == 0;
627 
628           // If there is a comma, semicolon or right paren after the closing
629           // brace, we assume this is a braced initializer list.  Note that
630           // regardless how we mark inner braces here, we will overwrite the
631           // BlockKind later if we parse a braced list (where all blocks
632           // inside are by default braced lists), or when we explicitly detect
633           // blocks (for example while parsing lambdas).
634           // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a
635           // braced list in JS.
636           ProbablyBracedList =
637               (Style.isJavaScript() &&
638                NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in,
639                                 Keywords.kw_as)) ||
640               (Style.isCpp() && NextTok->is(tok::l_paren)) ||
641               NextTok->isOneOf(tok::comma, tok::period, tok::colon,
642                                tok::r_paren, tok::r_square, tok::l_brace,
643                                tok::ellipsis) ||
644               (NextTok->is(tok::identifier) &&
645                !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) ||
646               (NextTok->is(tok::semi) &&
647                (!ExpectClassBody || LBraceStack.size() != 1)) ||
648               (NextTok->isBinaryOperator() && !NextIsObjCMethod);
649           if (!Style.isCSharp() && NextTok->is(tok::l_square)) {
650             // We can have an array subscript after a braced init
651             // list, but C++11 attributes are expected after blocks.
652             NextTok = Tokens->getNextToken();
653             ++ReadTokens;
654             ProbablyBracedList = NextTok->isNot(tok::l_square);
655           }
656         }
657         if (ProbablyBracedList) {
658           Tok->setBlockKind(BK_BracedInit);
659           LBraceStack.back()->setBlockKind(BK_BracedInit);
660         } else {
661           Tok->setBlockKind(BK_Block);
662           LBraceStack.back()->setBlockKind(BK_Block);
663         }
664       }
665       LBraceStack.pop_back();
666       break;
667     case tok::identifier:
668       if (!Tok->is(TT_StatementMacro))
669         break;
670       LLVM_FALLTHROUGH;
671     case tok::at:
672     case tok::semi:
673     case tok::kw_if:
674     case tok::kw_while:
675     case tok::kw_for:
676     case tok::kw_switch:
677     case tok::kw_try:
678     case tok::kw___try:
679       if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown))
680         LBraceStack.back()->setBlockKind(BK_Block);
681       break;
682     default:
683       break;
684     }
685     PrevTok = Tok;
686     Tok = NextTok;
687   } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty());
688 
689   // Assume other blocks for all unclosed opening braces.
690   for (FormatToken *LBrace : LBraceStack) {
691     if (LBrace->is(BK_Unknown))
692       LBrace->setBlockKind(BK_Block);
693   }
694 
695   FormatTok = Tokens->setPosition(StoredPosition);
696 }
697 
698 template <class T>
699 static inline void hash_combine(std::size_t &seed, const T &v) {
700   std::hash<T> hasher;
701   seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
702 }
703 
704 size_t UnwrappedLineParser::computePPHash() const {
705   size_t h = 0;
706   for (const auto &i : PPStack) {
707     hash_combine(h, size_t(i.Kind));
708     hash_combine(h, i.Line);
709   }
710   return h;
711 }
712 
713 UnwrappedLineParser::IfStmtKind
714 UnwrappedLineParser::parseBlock(bool MustBeDeclaration, unsigned AddLevels,
715                                 bool MunchSemi,
716                                 bool UnindentWhitesmithsBraces) {
717   assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) &&
718          "'{' or macro block token expected");
719   FormatToken *Tok = FormatTok;
720   const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin);
721   FormatTok->setBlockKind(BK_Block);
722 
723   // For Whitesmiths mode, jump to the next level prior to skipping over the
724   // braces.
725   if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
726     ++Line->Level;
727 
728   size_t PPStartHash = computePPHash();
729 
730   unsigned InitialLevel = Line->Level;
731   nextToken(/*LevelDifference=*/AddLevels);
732 
733   if (MacroBlock && FormatTok->is(tok::l_paren))
734     parseParens();
735 
736   size_t NbPreprocessorDirectives =
737       CurrentLines == &Lines ? PreprocessorDirectives.size() : 0;
738   addUnwrappedLine();
739   size_t OpeningLineIndex =
740       CurrentLines->empty()
741           ? (UnwrappedLine::kInvalidIndex)
742           : (CurrentLines->size() - 1 - NbPreprocessorDirectives);
743 
744   // Whitesmiths is weird here. The brace needs to be indented for the namespace
745   // block, but the block itself may not be indented depending on the style
746   // settings. This allows the format to back up one level in those cases.
747   if (UnindentWhitesmithsBraces)
748     --Line->Level;
749 
750   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
751                                           MustBeDeclaration);
752   if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths)
753     Line->Level += AddLevels;
754 
755   IfStmtKind IfKind = IfStmtKind::NotIf;
756   const bool SimpleBlock = parseLevel(/*HasOpeningBrace=*/true, &IfKind);
757 
758   if (eof())
759     return IfKind;
760 
761   if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd)
762                  : !FormatTok->is(tok::r_brace)) {
763     Line->Level = InitialLevel;
764     FormatTok->setBlockKind(BK_Block);
765     return IfKind;
766   }
767 
768   if (SimpleBlock && Tok->is(tok::l_brace)) {
769     assert(FormatTok->is(tok::r_brace));
770     const FormatToken *Previous = Tokens->getPreviousToken();
771     assert(Previous);
772     if (Previous->isNot(tok::r_brace) || Previous->Optional) {
773       Tok->MatchingParen = FormatTok;
774       FormatTok->MatchingParen = Tok;
775     }
776   }
777 
778   size_t PPEndHash = computePPHash();
779 
780   // Munch the closing brace.
781   nextToken(/*LevelDifference=*/-AddLevels);
782 
783   if (MacroBlock && FormatTok->is(tok::l_paren))
784     parseParens();
785 
786   if (FormatTok->is(tok::arrow)) {
787     // Following the } we can find a trailing return type arrow
788     // as part of an implicit conversion constraint.
789     nextToken();
790     parseStructuralElement();
791   }
792 
793   if (MunchSemi && FormatTok->Tok.is(tok::semi))
794     nextToken();
795 
796   Line->Level = InitialLevel;
797 
798   if (PPStartHash == PPEndHash) {
799     Line->MatchingOpeningBlockLineIndex = OpeningLineIndex;
800     if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) {
801       // Update the opening line to add the forward reference as well
802       (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex =
803           CurrentLines->size() - 1;
804     }
805   }
806 
807   return IfKind;
808 }
809 
810 static bool isGoogScope(const UnwrappedLine &Line) {
811   // FIXME: Closure-library specific stuff should not be hard-coded but be
812   // configurable.
813   if (Line.Tokens.size() < 4)
814     return false;
815   auto I = Line.Tokens.begin();
816   if (I->Tok->TokenText != "goog")
817     return false;
818   ++I;
819   if (I->Tok->isNot(tok::period))
820     return false;
821   ++I;
822   if (I->Tok->TokenText != "scope")
823     return false;
824   ++I;
825   return I->Tok->is(tok::l_paren);
826 }
827 
828 static bool isIIFE(const UnwrappedLine &Line,
829                    const AdditionalKeywords &Keywords) {
830   // Look for the start of an immediately invoked anonymous function.
831   // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression
832   // This is commonly done in JavaScript to create a new, anonymous scope.
833   // Example: (function() { ... })()
834   if (Line.Tokens.size() < 3)
835     return false;
836   auto I = Line.Tokens.begin();
837   if (I->Tok->isNot(tok::l_paren))
838     return false;
839   ++I;
840   if (I->Tok->isNot(Keywords.kw_function))
841     return false;
842   ++I;
843   return I->Tok->is(tok::l_paren);
844 }
845 
846 static bool ShouldBreakBeforeBrace(const FormatStyle &Style,
847                                    const FormatToken &InitialToken) {
848   if (InitialToken.isOneOf(tok::kw_namespace, TT_NamespaceMacro))
849     return Style.BraceWrapping.AfterNamespace;
850   if (InitialToken.is(tok::kw_class))
851     return Style.BraceWrapping.AfterClass;
852   if (InitialToken.is(tok::kw_union))
853     return Style.BraceWrapping.AfterUnion;
854   if (InitialToken.is(tok::kw_struct))
855     return Style.BraceWrapping.AfterStruct;
856   if (InitialToken.is(tok::kw_enum))
857     return Style.BraceWrapping.AfterEnum;
858   return false;
859 }
860 
861 void UnwrappedLineParser::parseChildBlock() {
862   FormatTok->setBlockKind(BK_Block);
863   nextToken();
864   {
865     bool SkipIndent = (Style.isJavaScript() &&
866                        (isGoogScope(*Line) || isIIFE(*Line, Keywords)));
867     ScopedLineState LineState(*this);
868     ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
869                                             /*MustBeDeclaration=*/false);
870     Line->Level += SkipIndent ? 0 : 1;
871     parseLevel(/*HasOpeningBrace=*/true);
872     flushComments(isOnNewLine(*FormatTok));
873     Line->Level -= SkipIndent ? 0 : 1;
874   }
875   nextToken();
876 }
877 
878 void UnwrappedLineParser::parsePPDirective() {
879   assert(FormatTok->Tok.is(tok::hash) && "'#' expected");
880   ScopedMacroState MacroState(*Line, Tokens, FormatTok);
881 
882   nextToken();
883 
884   if (!FormatTok->Tok.getIdentifierInfo()) {
885     parsePPUnknown();
886     return;
887   }
888 
889   switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) {
890   case tok::pp_define:
891     parsePPDefine();
892     return;
893   case tok::pp_if:
894     parsePPIf(/*IfDef=*/false);
895     break;
896   case tok::pp_ifdef:
897   case tok::pp_ifndef:
898     parsePPIf(/*IfDef=*/true);
899     break;
900   case tok::pp_else:
901     parsePPElse();
902     break;
903   case tok::pp_elifdef:
904   case tok::pp_elifndef:
905   case tok::pp_elif:
906     parsePPElIf();
907     break;
908   case tok::pp_endif:
909     parsePPEndIf();
910     break;
911   default:
912     parsePPUnknown();
913     break;
914   }
915 }
916 
917 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) {
918   size_t Line = CurrentLines->size();
919   if (CurrentLines == &PreprocessorDirectives)
920     Line += Lines.size();
921 
922   if (Unreachable ||
923       (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable))
924     PPStack.push_back({PP_Unreachable, Line});
925   else
926     PPStack.push_back({PP_Conditional, Line});
927 }
928 
929 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) {
930   ++PPBranchLevel;
931   assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size());
932   if (PPBranchLevel == (int)PPLevelBranchIndex.size()) {
933     PPLevelBranchIndex.push_back(0);
934     PPLevelBranchCount.push_back(0);
935   }
936   PPChainBranchIndex.push(0);
937   bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0;
938   conditionalCompilationCondition(Unreachable || Skip);
939 }
940 
941 void UnwrappedLineParser::conditionalCompilationAlternative() {
942   if (!PPStack.empty())
943     PPStack.pop_back();
944   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
945   if (!PPChainBranchIndex.empty())
946     ++PPChainBranchIndex.top();
947   conditionalCompilationCondition(
948       PPBranchLevel >= 0 && !PPChainBranchIndex.empty() &&
949       PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top());
950 }
951 
952 void UnwrappedLineParser::conditionalCompilationEnd() {
953   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
954   if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) {
955     if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) {
956       PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1;
957     }
958   }
959   // Guard against #endif's without #if.
960   if (PPBranchLevel > -1)
961     --PPBranchLevel;
962   if (!PPChainBranchIndex.empty())
963     PPChainBranchIndex.pop();
964   if (!PPStack.empty())
965     PPStack.pop_back();
966 }
967 
968 void UnwrappedLineParser::parsePPIf(bool IfDef) {
969   bool IfNDef = FormatTok->is(tok::pp_ifndef);
970   nextToken();
971   bool Unreachable = false;
972   if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0"))
973     Unreachable = true;
974   if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG")
975     Unreachable = true;
976   conditionalCompilationStart(Unreachable);
977   FormatToken *IfCondition = FormatTok;
978   // If there's a #ifndef on the first line, and the only lines before it are
979   // comments, it could be an include guard.
980   bool MaybeIncludeGuard = IfNDef;
981   if (IncludeGuard == IG_Inited && MaybeIncludeGuard)
982     for (auto &Line : Lines) {
983       if (!Line.Tokens.front().Tok->is(tok::comment)) {
984         MaybeIncludeGuard = false;
985         IncludeGuard = IG_Rejected;
986         break;
987       }
988     }
989   --PPBranchLevel;
990   parsePPUnknown();
991   ++PPBranchLevel;
992   if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
993     IncludeGuard = IG_IfNdefed;
994     IncludeGuardToken = IfCondition;
995   }
996 }
997 
998 void UnwrappedLineParser::parsePPElse() {
999   // If a potential include guard has an #else, it's not an include guard.
1000   if (IncludeGuard == IG_Defined && PPBranchLevel == 0)
1001     IncludeGuard = IG_Rejected;
1002   conditionalCompilationAlternative();
1003   if (PPBranchLevel > -1)
1004     --PPBranchLevel;
1005   parsePPUnknown();
1006   ++PPBranchLevel;
1007 }
1008 
1009 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); }
1010 
1011 void UnwrappedLineParser::parsePPEndIf() {
1012   conditionalCompilationEnd();
1013   parsePPUnknown();
1014   // If the #endif of a potential include guard is the last thing in the file,
1015   // then we found an include guard.
1016   if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() &&
1017       Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1018     IncludeGuard = IG_Found;
1019 }
1020 
1021 void UnwrappedLineParser::parsePPDefine() {
1022   nextToken();
1023 
1024   if (!FormatTok->Tok.getIdentifierInfo()) {
1025     IncludeGuard = IG_Rejected;
1026     IncludeGuardToken = nullptr;
1027     parsePPUnknown();
1028     return;
1029   }
1030 
1031   if (IncludeGuard == IG_IfNdefed &&
1032       IncludeGuardToken->TokenText == FormatTok->TokenText) {
1033     IncludeGuard = IG_Defined;
1034     IncludeGuardToken = nullptr;
1035     for (auto &Line : Lines) {
1036       if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) {
1037         IncludeGuard = IG_Rejected;
1038         break;
1039       }
1040     }
1041   }
1042 
1043   nextToken();
1044   if (FormatTok->Tok.getKind() == tok::l_paren &&
1045       !FormatTok->hasWhitespaceBefore()) {
1046     parseParens();
1047   }
1048   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1049     Line->Level += PPBranchLevel + 1;
1050   addUnwrappedLine();
1051   ++Line->Level;
1052 
1053   // Errors during a preprocessor directive can only affect the layout of the
1054   // preprocessor directive, and thus we ignore them. An alternative approach
1055   // would be to use the same approach we use on the file level (no
1056   // re-indentation if there was a structural error) within the macro
1057   // definition.
1058   parseFile();
1059 }
1060 
1061 void UnwrappedLineParser::parsePPUnknown() {
1062   do {
1063     nextToken();
1064   } while (!eof());
1065   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1066     Line->Level += PPBranchLevel + 1;
1067   addUnwrappedLine();
1068 }
1069 
1070 // Here we exclude certain tokens that are not usually the first token in an
1071 // unwrapped line. This is used in attempt to distinguish macro calls without
1072 // trailing semicolons from other constructs split to several lines.
1073 static bool tokenCanStartNewLine(const FormatToken &Tok) {
1074   // Semicolon can be a null-statement, l_square can be a start of a macro or
1075   // a C++11 attribute, but this doesn't seem to be common.
1076   return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) &&
1077          Tok.isNot(TT_AttributeSquare) &&
1078          // Tokens that can only be used as binary operators and a part of
1079          // overloaded operator names.
1080          Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) &&
1081          Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) &&
1082          Tok.isNot(tok::less) && Tok.isNot(tok::greater) &&
1083          Tok.isNot(tok::slash) && Tok.isNot(tok::percent) &&
1084          Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) &&
1085          Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) &&
1086          Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) &&
1087          Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) &&
1088          Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) &&
1089          Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) &&
1090          Tok.isNot(tok::lesslessequal) &&
1091          // Colon is used in labels, base class lists, initializer lists,
1092          // range-based for loops, ternary operator, but should never be the
1093          // first token in an unwrapped line.
1094          Tok.isNot(tok::colon) &&
1095          // 'noexcept' is a trailing annotation.
1096          Tok.isNot(tok::kw_noexcept);
1097 }
1098 
1099 static bool mustBeJSIdent(const AdditionalKeywords &Keywords,
1100                           const FormatToken *FormatTok) {
1101   // FIXME: This returns true for C/C++ keywords like 'struct'.
1102   return FormatTok->is(tok::identifier) &&
1103          (FormatTok->Tok.getIdentifierInfo() == nullptr ||
1104           !FormatTok->isOneOf(
1105               Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async,
1106               Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally,
1107               Keywords.kw_function, Keywords.kw_import, Keywords.kw_is,
1108               Keywords.kw_let, Keywords.kw_var, tok::kw_const,
1109               Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements,
1110               Keywords.kw_instanceof, Keywords.kw_interface,
1111               Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from));
1112 }
1113 
1114 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords,
1115                                  const FormatToken *FormatTok) {
1116   return FormatTok->Tok.isLiteral() ||
1117          FormatTok->isOneOf(tok::kw_true, tok::kw_false) ||
1118          mustBeJSIdent(Keywords, FormatTok);
1119 }
1120 
1121 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement
1122 // when encountered after a value (see mustBeJSIdentOrValue).
1123 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords,
1124                            const FormatToken *FormatTok) {
1125   return FormatTok->isOneOf(
1126       tok::kw_return, Keywords.kw_yield,
1127       // conditionals
1128       tok::kw_if, tok::kw_else,
1129       // loops
1130       tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break,
1131       // switch/case
1132       tok::kw_switch, tok::kw_case,
1133       // exceptions
1134       tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally,
1135       // declaration
1136       tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let,
1137       Keywords.kw_async, Keywords.kw_function,
1138       // import/export
1139       Keywords.kw_import, tok::kw_export);
1140 }
1141 
1142 // Checks whether a token is a type in K&R C (aka C78).
1143 static bool isC78Type(const FormatToken &Tok) {
1144   return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long,
1145                      tok::kw_unsigned, tok::kw_float, tok::kw_double,
1146                      tok::identifier);
1147 }
1148 
1149 // This function checks whether a token starts the first parameter declaration
1150 // in a K&R C (aka C78) function definition, e.g.:
1151 //   int f(a, b)
1152 //   short a, b;
1153 //   {
1154 //      return a + b;
1155 //   }
1156 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next,
1157                                const FormatToken *FuncName) {
1158   assert(Tok);
1159   assert(Next);
1160   assert(FuncName);
1161 
1162   if (FuncName->isNot(tok::identifier))
1163     return false;
1164 
1165   const FormatToken *Prev = FuncName->Previous;
1166   if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev)))
1167     return false;
1168 
1169   if (!isC78Type(*Tok) &&
1170       !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union))
1171     return false;
1172 
1173   if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo())
1174     return false;
1175 
1176   Tok = Tok->Previous;
1177   if (!Tok || Tok->isNot(tok::r_paren))
1178     return false;
1179 
1180   Tok = Tok->Previous;
1181   if (!Tok || Tok->isNot(tok::identifier))
1182     return false;
1183 
1184   return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma);
1185 }
1186 
1187 void UnwrappedLineParser::parseModuleImport() {
1188   nextToken();
1189   while (!eof()) {
1190     if (FormatTok->is(tok::colon)) {
1191       FormatTok->setType(TT_ModulePartitionColon);
1192     }
1193     // Handle import <foo/bar.h> as we would an include statement.
1194     else if (FormatTok->is(tok::less)) {
1195       nextToken();
1196       while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) {
1197         // Mark tokens up to the trailing line comments as implicit string
1198         // literals.
1199         if (FormatTok->isNot(tok::comment) &&
1200             !FormatTok->TokenText.startswith("//"))
1201           FormatTok->setType(TT_ImplicitStringLiteral);
1202         nextToken();
1203       }
1204     }
1205     if (FormatTok->is(tok::semi)) {
1206       nextToken();
1207       break;
1208     }
1209     nextToken();
1210   }
1211 
1212   addUnwrappedLine();
1213 }
1214 
1215 // readTokenWithJavaScriptASI reads the next token and terminates the current
1216 // line if JavaScript Automatic Semicolon Insertion must
1217 // happen between the current token and the next token.
1218 //
1219 // This method is conservative - it cannot cover all edge cases of JavaScript,
1220 // but only aims to correctly handle certain well known cases. It *must not*
1221 // return true in speculative cases.
1222 void UnwrappedLineParser::readTokenWithJavaScriptASI() {
1223   FormatToken *Previous = FormatTok;
1224   readToken();
1225   FormatToken *Next = FormatTok;
1226 
1227   bool IsOnSameLine =
1228       CommentsBeforeNextToken.empty()
1229           ? Next->NewlinesBefore == 0
1230           : CommentsBeforeNextToken.front()->NewlinesBefore == 0;
1231   if (IsOnSameLine)
1232     return;
1233 
1234   bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous);
1235   bool PreviousStartsTemplateExpr =
1236       Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${");
1237   if (PreviousMustBeValue || Previous->is(tok::r_paren)) {
1238     // If the line contains an '@' sign, the previous token might be an
1239     // annotation, which can precede another identifier/value.
1240     bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) {
1241       return LineNode.Tok->is(tok::at);
1242     });
1243     if (HasAt)
1244       return;
1245   }
1246   if (Next->is(tok::exclaim) && PreviousMustBeValue)
1247     return addUnwrappedLine();
1248   bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next);
1249   bool NextEndsTemplateExpr =
1250       Next->is(TT_TemplateString) && Next->TokenText.startswith("}");
1251   if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr &&
1252       (PreviousMustBeValue ||
1253        Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus,
1254                          tok::minusminus)))
1255     return addUnwrappedLine();
1256   if ((PreviousMustBeValue || Previous->is(tok::r_paren)) &&
1257       isJSDeclOrStmt(Keywords, Next))
1258     return addUnwrappedLine();
1259 }
1260 
1261 void UnwrappedLineParser::parseStructuralElement(IfStmtKind *IfKind,
1262                                                  bool IsTopLevel) {
1263   if (Style.Language == FormatStyle::LK_TableGen &&
1264       FormatTok->is(tok::pp_include)) {
1265     nextToken();
1266     if (FormatTok->is(tok::string_literal))
1267       nextToken();
1268     addUnwrappedLine();
1269     return;
1270   }
1271   switch (FormatTok->Tok.getKind()) {
1272   case tok::kw_asm:
1273     nextToken();
1274     if (FormatTok->is(tok::l_brace)) {
1275       FormatTok->setType(TT_InlineASMBrace);
1276       nextToken();
1277       while (FormatTok && FormatTok->isNot(tok::eof)) {
1278         if (FormatTok->is(tok::r_brace)) {
1279           FormatTok->setType(TT_InlineASMBrace);
1280           nextToken();
1281           addUnwrappedLine();
1282           break;
1283         }
1284         FormatTok->Finalized = true;
1285         nextToken();
1286       }
1287     }
1288     break;
1289   case tok::kw_namespace:
1290     parseNamespace();
1291     return;
1292   case tok::kw_public:
1293   case tok::kw_protected:
1294   case tok::kw_private:
1295     if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
1296         Style.isCSharp())
1297       nextToken();
1298     else
1299       parseAccessSpecifier();
1300     return;
1301   case tok::kw_if:
1302     if (Style.isJavaScript() && Line->MustBeDeclaration)
1303       // field/method declaration.
1304       break;
1305     parseIfThenElse(IfKind);
1306     return;
1307   case tok::kw_for:
1308   case tok::kw_while:
1309     if (Style.isJavaScript() && Line->MustBeDeclaration)
1310       // field/method declaration.
1311       break;
1312     parseForOrWhileLoop();
1313     return;
1314   case tok::kw_do:
1315     if (Style.isJavaScript() && Line->MustBeDeclaration)
1316       // field/method declaration.
1317       break;
1318     parseDoWhile();
1319     return;
1320   case tok::kw_switch:
1321     if (Style.isJavaScript() && Line->MustBeDeclaration)
1322       // 'switch: string' field declaration.
1323       break;
1324     parseSwitch();
1325     return;
1326   case tok::kw_default:
1327     if (Style.isJavaScript() && Line->MustBeDeclaration)
1328       // 'default: string' field declaration.
1329       break;
1330     nextToken();
1331     if (FormatTok->is(tok::colon)) {
1332       parseLabel();
1333       return;
1334     }
1335     // e.g. "default void f() {}" in a Java interface.
1336     break;
1337   case tok::kw_case:
1338     if (Style.isJavaScript() && Line->MustBeDeclaration)
1339       // 'case: string' field declaration.
1340       break;
1341     parseCaseLabel();
1342     return;
1343   case tok::kw_try:
1344   case tok::kw___try:
1345     if (Style.isJavaScript() && Line->MustBeDeclaration)
1346       // field/method declaration.
1347       break;
1348     parseTryCatch();
1349     return;
1350   case tok::kw_extern:
1351     nextToken();
1352     if (FormatTok->Tok.is(tok::string_literal)) {
1353       nextToken();
1354       if (FormatTok->Tok.is(tok::l_brace)) {
1355         if (Style.BraceWrapping.AfterExternBlock)
1356           addUnwrappedLine();
1357         // Either we indent or for backwards compatibility we follow the
1358         // AfterExternBlock style.
1359         unsigned AddLevels =
1360             (Style.IndentExternBlock == FormatStyle::IEBS_Indent) ||
1361                     (Style.BraceWrapping.AfterExternBlock &&
1362                      Style.IndentExternBlock ==
1363                          FormatStyle::IEBS_AfterExternBlock)
1364                 ? 1u
1365                 : 0u;
1366         parseBlock(/*MustBeDeclaration=*/true, AddLevels);
1367         addUnwrappedLine();
1368         return;
1369       }
1370     }
1371     break;
1372   case tok::kw_export:
1373     if (Style.isJavaScript()) {
1374       parseJavaScriptEs6ImportExport();
1375       return;
1376     }
1377     if (!Style.isCpp())
1378       break;
1379     // Handle C++ "(inline|export) namespace".
1380     LLVM_FALLTHROUGH;
1381   case tok::kw_inline:
1382     nextToken();
1383     if (FormatTok->Tok.is(tok::kw_namespace)) {
1384       parseNamespace();
1385       return;
1386     }
1387     break;
1388   case tok::identifier:
1389     if (FormatTok->is(TT_ForEachMacro)) {
1390       parseForOrWhileLoop();
1391       return;
1392     }
1393     if (FormatTok->is(TT_MacroBlockBegin)) {
1394       parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
1395                  /*MunchSemi=*/false);
1396       return;
1397     }
1398     if (FormatTok->is(Keywords.kw_import)) {
1399       if (Style.isJavaScript()) {
1400         parseJavaScriptEs6ImportExport();
1401         return;
1402       }
1403       if (Style.Language == FormatStyle::LK_Proto) {
1404         nextToken();
1405         if (FormatTok->is(tok::kw_public))
1406           nextToken();
1407         if (!FormatTok->is(tok::string_literal))
1408           return;
1409         nextToken();
1410         if (FormatTok->is(tok::semi))
1411           nextToken();
1412         addUnwrappedLine();
1413         return;
1414       }
1415       if (Style.isCpp()) {
1416         parseModuleImport();
1417         return;
1418       }
1419     }
1420     if (Style.isCpp() &&
1421         FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals,
1422                            Keywords.kw_slots, Keywords.kw_qslots)) {
1423       nextToken();
1424       if (FormatTok->is(tok::colon)) {
1425         nextToken();
1426         addUnwrappedLine();
1427         return;
1428       }
1429     }
1430     if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) {
1431       parseStatementMacro();
1432       return;
1433     }
1434     if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) {
1435       parseNamespace();
1436       return;
1437     }
1438     // In all other cases, parse the declaration.
1439     break;
1440   default:
1441     break;
1442   }
1443   do {
1444     const FormatToken *Previous = FormatTok->Previous;
1445     switch (FormatTok->Tok.getKind()) {
1446     case tok::at:
1447       nextToken();
1448       if (FormatTok->Tok.is(tok::l_brace)) {
1449         nextToken();
1450         parseBracedList();
1451         break;
1452       } else if (Style.Language == FormatStyle::LK_Java &&
1453                  FormatTok->is(Keywords.kw_interface)) {
1454         nextToken();
1455         break;
1456       }
1457       switch (FormatTok->Tok.getObjCKeywordID()) {
1458       case tok::objc_public:
1459       case tok::objc_protected:
1460       case tok::objc_package:
1461       case tok::objc_private:
1462         return parseAccessSpecifier();
1463       case tok::objc_interface:
1464       case tok::objc_implementation:
1465         return parseObjCInterfaceOrImplementation();
1466       case tok::objc_protocol:
1467         if (parseObjCProtocol())
1468           return;
1469         break;
1470       case tok::objc_end:
1471         return; // Handled by the caller.
1472       case tok::objc_optional:
1473       case tok::objc_required:
1474         nextToken();
1475         addUnwrappedLine();
1476         return;
1477       case tok::objc_autoreleasepool:
1478         nextToken();
1479         if (FormatTok->Tok.is(tok::l_brace)) {
1480           if (Style.BraceWrapping.AfterControlStatement ==
1481               FormatStyle::BWACS_Always)
1482             addUnwrappedLine();
1483           parseBlock();
1484         }
1485         addUnwrappedLine();
1486         return;
1487       case tok::objc_synchronized:
1488         nextToken();
1489         if (FormatTok->Tok.is(tok::l_paren))
1490           // Skip synchronization object
1491           parseParens();
1492         if (FormatTok->Tok.is(tok::l_brace)) {
1493           if (Style.BraceWrapping.AfterControlStatement ==
1494               FormatStyle::BWACS_Always)
1495             addUnwrappedLine();
1496           parseBlock();
1497         }
1498         addUnwrappedLine();
1499         return;
1500       case tok::objc_try:
1501         // This branch isn't strictly necessary (the kw_try case below would
1502         // do this too after the tok::at is parsed above).  But be explicit.
1503         parseTryCatch();
1504         return;
1505       default:
1506         break;
1507       }
1508       break;
1509     case tok::kw_concept:
1510       parseConcept();
1511       return;
1512     case tok::kw_requires:
1513       parseRequires();
1514       return;
1515     case tok::kw_enum:
1516       // Ignore if this is part of "template <enum ...".
1517       if (Previous && Previous->is(tok::less)) {
1518         nextToken();
1519         break;
1520       }
1521 
1522       // parseEnum falls through and does not yet add an unwrapped line as an
1523       // enum definition can start a structural element.
1524       if (!parseEnum())
1525         break;
1526       // This only applies for C++.
1527       if (!Style.isCpp()) {
1528         addUnwrappedLine();
1529         return;
1530       }
1531       break;
1532     case tok::kw_typedef:
1533       nextToken();
1534       if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS,
1535                              Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS,
1536                              Keywords.kw_CF_CLOSED_ENUM,
1537                              Keywords.kw_NS_CLOSED_ENUM))
1538         parseEnum();
1539       break;
1540     case tok::kw_struct:
1541     case tok::kw_union:
1542     case tok::kw_class:
1543       if (parseStructLike()) {
1544         return;
1545       }
1546       break;
1547     case tok::period:
1548       nextToken();
1549       // In Java, classes have an implicit static member "class".
1550       if (Style.Language == FormatStyle::LK_Java && FormatTok &&
1551           FormatTok->is(tok::kw_class))
1552         nextToken();
1553       if (Style.isJavaScript() && FormatTok &&
1554           FormatTok->Tok.getIdentifierInfo())
1555         // JavaScript only has pseudo keywords, all keywords are allowed to
1556         // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6
1557         nextToken();
1558       break;
1559     case tok::semi:
1560       nextToken();
1561       addUnwrappedLine();
1562       return;
1563     case tok::r_brace:
1564       addUnwrappedLine();
1565       return;
1566     case tok::l_paren: {
1567       parseParens();
1568       // Break the unwrapped line if a K&R C function definition has a parameter
1569       // declaration.
1570       if (!IsTopLevel || !Style.isCpp() || !Previous || FormatTok->is(tok::eof))
1571         break;
1572       if (isC78ParameterDecl(FormatTok, Tokens->peekNextToken(), Previous)) {
1573         addUnwrappedLine();
1574         return;
1575       }
1576       break;
1577     }
1578     case tok::kw_operator:
1579       nextToken();
1580       if (FormatTok->isBinaryOperator())
1581         nextToken();
1582       break;
1583     case tok::caret:
1584       nextToken();
1585       if (FormatTok->Tok.isAnyIdentifier() ||
1586           FormatTok->isSimpleTypeSpecifier())
1587         nextToken();
1588       if (FormatTok->is(tok::l_paren))
1589         parseParens();
1590       if (FormatTok->is(tok::l_brace))
1591         parseChildBlock();
1592       break;
1593     case tok::l_brace:
1594       if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) {
1595         // A block outside of parentheses must be the last part of a
1596         // structural element.
1597         // FIXME: Figure out cases where this is not true, and add projections
1598         // for them (the one we know is missing are lambdas).
1599         if (Style.Language == FormatStyle::LK_Java &&
1600             Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) {
1601           // If necessary, we could set the type to something different than
1602           // TT_FunctionLBrace.
1603           if (Style.BraceWrapping.AfterControlStatement ==
1604               FormatStyle::BWACS_Always)
1605             addUnwrappedLine();
1606         } else if (Style.BraceWrapping.AfterFunction) {
1607           addUnwrappedLine();
1608         }
1609         FormatTok->setType(TT_FunctionLBrace);
1610         parseBlock();
1611         addUnwrappedLine();
1612         return;
1613       }
1614       // Otherwise this was a braced init list, and the structural
1615       // element continues.
1616       break;
1617     case tok::kw_try:
1618       if (Style.isJavaScript() && Line->MustBeDeclaration) {
1619         // field/method declaration.
1620         nextToken();
1621         break;
1622       }
1623       // We arrive here when parsing function-try blocks.
1624       if (Style.BraceWrapping.AfterFunction)
1625         addUnwrappedLine();
1626       parseTryCatch();
1627       return;
1628     case tok::identifier: {
1629       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) &&
1630           Line->MustBeDeclaration) {
1631         addUnwrappedLine();
1632         parseCSharpGenericTypeConstraint();
1633         break;
1634       }
1635       if (FormatTok->is(TT_MacroBlockEnd)) {
1636         addUnwrappedLine();
1637         return;
1638       }
1639 
1640       // Function declarations (as opposed to function expressions) are parsed
1641       // on their own unwrapped line by continuing this loop. Function
1642       // expressions (functions that are not on their own line) must not create
1643       // a new unwrapped line, so they are special cased below.
1644       size_t TokenCount = Line->Tokens.size();
1645       if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) &&
1646           (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is(
1647                                                      Keywords.kw_async)))) {
1648         tryToParseJSFunction();
1649         break;
1650       }
1651       if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) &&
1652           FormatTok->is(Keywords.kw_interface)) {
1653         if (Style.isJavaScript()) {
1654           // In JavaScript/TypeScript, "interface" can be used as a standalone
1655           // identifier, e.g. in `var interface = 1;`. If "interface" is
1656           // followed by another identifier, it is very like to be an actual
1657           // interface declaration.
1658           unsigned StoredPosition = Tokens->getPosition();
1659           FormatToken *Next = Tokens->getNextToken();
1660           FormatTok = Tokens->setPosition(StoredPosition);
1661           if (!mustBeJSIdent(Keywords, Next)) {
1662             nextToken();
1663             break;
1664           }
1665         }
1666         parseRecord();
1667         addUnwrappedLine();
1668         return;
1669       }
1670 
1671       if (FormatTok->is(Keywords.kw_interface)) {
1672         if (parseStructLike()) {
1673           return;
1674         }
1675         break;
1676       }
1677 
1678       if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) {
1679         parseStatementMacro();
1680         return;
1681       }
1682 
1683       // See if the following token should start a new unwrapped line.
1684       StringRef Text = FormatTok->TokenText;
1685 
1686       FormatToken *PreviousToken = FormatTok;
1687       nextToken();
1688 
1689       // JS doesn't have macros, and within classes colons indicate fields, not
1690       // labels.
1691       if (Style.isJavaScript())
1692         break;
1693 
1694       TokenCount = Line->Tokens.size();
1695       if (TokenCount == 1 ||
1696           (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) {
1697         if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) {
1698           Line->Tokens.begin()->Tok->MustBreakBefore = true;
1699           parseLabel(!Style.IndentGotoLabels);
1700           return;
1701         }
1702         // Recognize function-like macro usages without trailing semicolon as
1703         // well as free-standing macros like Q_OBJECT.
1704         bool FunctionLike = FormatTok->is(tok::l_paren);
1705         if (FunctionLike)
1706           parseParens();
1707 
1708         bool FollowedByNewline =
1709             CommentsBeforeNextToken.empty()
1710                 ? FormatTok->NewlinesBefore > 0
1711                 : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
1712 
1713         if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) &&
1714             tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) {
1715           PreviousToken->setType(TT_FunctionLikeOrFreestandingMacro);
1716           addUnwrappedLine();
1717           return;
1718         }
1719       }
1720       break;
1721     }
1722     case tok::equal:
1723       if ((Style.isJavaScript() || Style.isCSharp()) &&
1724           FormatTok->is(TT_FatArrow)) {
1725         tryToParseChildBlock();
1726         break;
1727       }
1728 
1729       nextToken();
1730       if (FormatTok->Tok.is(tok::l_brace)) {
1731         // Block kind should probably be set to BK_BracedInit for any language.
1732         // C# needs this change to ensure that array initialisers and object
1733         // initialisers are indented the same way.
1734         if (Style.isCSharp())
1735           FormatTok->setBlockKind(BK_BracedInit);
1736         nextToken();
1737         parseBracedList();
1738       } else if (Style.Language == FormatStyle::LK_Proto &&
1739                  FormatTok->Tok.is(tok::less)) {
1740         nextToken();
1741         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
1742                         /*ClosingBraceKind=*/tok::greater);
1743       }
1744       break;
1745     case tok::l_square:
1746       parseSquare();
1747       break;
1748     case tok::kw_new:
1749       parseNew();
1750       break;
1751     default:
1752       nextToken();
1753       break;
1754     }
1755   } while (!eof());
1756 }
1757 
1758 bool UnwrappedLineParser::tryToParsePropertyAccessor() {
1759   assert(FormatTok->is(tok::l_brace));
1760   if (!Style.isCSharp())
1761     return false;
1762   // See if it's a property accessor.
1763   if (FormatTok->Previous->isNot(tok::identifier))
1764     return false;
1765 
1766   // See if we are inside a property accessor.
1767   //
1768   // Record the current tokenPosition so that we can advance and
1769   // reset the current token. `Next` is not set yet so we need
1770   // another way to advance along the token stream.
1771   unsigned int StoredPosition = Tokens->getPosition();
1772   FormatToken *Tok = Tokens->getNextToken();
1773 
1774   // A trivial property accessor is of the form:
1775   // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] }
1776   // Track these as they do not require line breaks to be introduced.
1777   bool HasGetOrSet = false;
1778   bool IsTrivialPropertyAccessor = true;
1779   while (!eof()) {
1780     if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private,
1781                      tok::kw_protected, Keywords.kw_internal, Keywords.kw_get,
1782                      Keywords.kw_set)) {
1783       if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set))
1784         HasGetOrSet = true;
1785       Tok = Tokens->getNextToken();
1786       continue;
1787     }
1788     if (Tok->isNot(tok::r_brace))
1789       IsTrivialPropertyAccessor = false;
1790     break;
1791   }
1792 
1793   if (!HasGetOrSet) {
1794     Tokens->setPosition(StoredPosition);
1795     return false;
1796   }
1797 
1798   // Try to parse the property accessor:
1799   // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties
1800   Tokens->setPosition(StoredPosition);
1801   if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction)
1802     addUnwrappedLine();
1803   nextToken();
1804   do {
1805     switch (FormatTok->Tok.getKind()) {
1806     case tok::r_brace:
1807       nextToken();
1808       if (FormatTok->is(tok::equal)) {
1809         while (!eof() && FormatTok->isNot(tok::semi))
1810           nextToken();
1811         nextToken();
1812       }
1813       addUnwrappedLine();
1814       return true;
1815     case tok::l_brace:
1816       ++Line->Level;
1817       parseBlock(/*MustBeDeclaration=*/true);
1818       addUnwrappedLine();
1819       --Line->Level;
1820       break;
1821     case tok::equal:
1822       if (FormatTok->is(TT_FatArrow)) {
1823         ++Line->Level;
1824         do {
1825           nextToken();
1826         } while (!eof() && FormatTok->isNot(tok::semi));
1827         nextToken();
1828         addUnwrappedLine();
1829         --Line->Level;
1830         break;
1831       }
1832       nextToken();
1833       break;
1834     default:
1835       if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) &&
1836           !IsTrivialPropertyAccessor) {
1837         // Non-trivial get/set needs to be on its own line.
1838         addUnwrappedLine();
1839       }
1840       nextToken();
1841     }
1842   } while (!eof());
1843 
1844   // Unreachable for well-formed code (paired '{' and '}').
1845   return true;
1846 }
1847 
1848 bool UnwrappedLineParser::tryToParseLambda() {
1849   if (!Style.isCpp()) {
1850     nextToken();
1851     return false;
1852   }
1853   assert(FormatTok->is(tok::l_square));
1854   FormatToken &LSquare = *FormatTok;
1855   if (!tryToParseLambdaIntroducer())
1856     return false;
1857 
1858   bool SeenArrow = false;
1859   bool InTemplateParameterList = false;
1860 
1861   while (FormatTok->isNot(tok::l_brace)) {
1862     if (FormatTok->isSimpleTypeSpecifier()) {
1863       nextToken();
1864       continue;
1865     }
1866     switch (FormatTok->Tok.getKind()) {
1867     case tok::l_brace:
1868       break;
1869     case tok::l_paren:
1870       parseParens();
1871       break;
1872     case tok::l_square:
1873       parseSquare();
1874       break;
1875     case tok::kw_class:
1876     case tok::kw_template:
1877     case tok::kw_typename:
1878       assert(FormatTok->Previous);
1879       if (FormatTok->Previous->is(tok::less))
1880         InTemplateParameterList = true;
1881       nextToken();
1882       break;
1883     case tok::amp:
1884     case tok::star:
1885     case tok::kw_const:
1886     case tok::comma:
1887     case tok::less:
1888     case tok::greater:
1889     case tok::identifier:
1890     case tok::numeric_constant:
1891     case tok::coloncolon:
1892     case tok::kw_mutable:
1893     case tok::kw_noexcept:
1894       nextToken();
1895       break;
1896     // Specialization of a template with an integer parameter can contain
1897     // arithmetic, logical, comparison and ternary operators.
1898     //
1899     // FIXME: This also accepts sequences of operators that are not in the scope
1900     // of a template argument list.
1901     //
1902     // In a C++ lambda a template type can only occur after an arrow. We use
1903     // this as an heuristic to distinguish between Objective-C expressions
1904     // followed by an `a->b` expression, such as:
1905     // ([obj func:arg] + a->b)
1906     // Otherwise the code below would parse as a lambda.
1907     //
1908     // FIXME: This heuristic is incorrect for C++20 generic lambdas with
1909     // explicit template lists: []<bool b = true && false>(U &&u){}
1910     case tok::plus:
1911     case tok::minus:
1912     case tok::exclaim:
1913     case tok::tilde:
1914     case tok::slash:
1915     case tok::percent:
1916     case tok::lessless:
1917     case tok::pipe:
1918     case tok::pipepipe:
1919     case tok::ampamp:
1920     case tok::caret:
1921     case tok::equalequal:
1922     case tok::exclaimequal:
1923     case tok::greaterequal:
1924     case tok::lessequal:
1925     case tok::question:
1926     case tok::colon:
1927     case tok::ellipsis:
1928     case tok::kw_true:
1929     case tok::kw_false:
1930       if (SeenArrow || InTemplateParameterList) {
1931         nextToken();
1932         break;
1933       }
1934       return true;
1935     case tok::arrow:
1936       // This might or might not actually be a lambda arrow (this could be an
1937       // ObjC method invocation followed by a dereferencing arrow). We might
1938       // reset this back to TT_Unknown in TokenAnnotator.
1939       FormatTok->setType(TT_LambdaArrow);
1940       SeenArrow = true;
1941       nextToken();
1942       break;
1943     default:
1944       return true;
1945     }
1946   }
1947   FormatTok->setType(TT_LambdaLBrace);
1948   LSquare.setType(TT_LambdaLSquare);
1949   parseChildBlock();
1950   return true;
1951 }
1952 
1953 bool UnwrappedLineParser::tryToParseLambdaIntroducer() {
1954   const FormatToken *Previous = FormatTok->Previous;
1955   if (Previous &&
1956       (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new,
1957                          tok::kw_delete, tok::l_square) ||
1958        FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() ||
1959        Previous->isSimpleTypeSpecifier())) {
1960     nextToken();
1961     return false;
1962   }
1963   nextToken();
1964   if (FormatTok->is(tok::l_square)) {
1965     return false;
1966   }
1967   parseSquare(/*LambdaIntroducer=*/true);
1968   return true;
1969 }
1970 
1971 void UnwrappedLineParser::tryToParseJSFunction() {
1972   assert(FormatTok->is(Keywords.kw_function) ||
1973          FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function));
1974   if (FormatTok->is(Keywords.kw_async))
1975     nextToken();
1976   // Consume "function".
1977   nextToken();
1978 
1979   // Consume * (generator function). Treat it like C++'s overloaded operators.
1980   if (FormatTok->is(tok::star)) {
1981     FormatTok->setType(TT_OverloadedOperator);
1982     nextToken();
1983   }
1984 
1985   // Consume function name.
1986   if (FormatTok->is(tok::identifier))
1987     nextToken();
1988 
1989   if (FormatTok->isNot(tok::l_paren))
1990     return;
1991 
1992   // Parse formal parameter list.
1993   parseParens();
1994 
1995   if (FormatTok->is(tok::colon)) {
1996     // Parse a type definition.
1997     nextToken();
1998 
1999     // Eat the type declaration. For braced inline object types, balance braces,
2000     // otherwise just parse until finding an l_brace for the function body.
2001     if (FormatTok->is(tok::l_brace))
2002       tryToParseBracedList();
2003     else
2004       while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof())
2005         nextToken();
2006   }
2007 
2008   if (FormatTok->is(tok::semi))
2009     return;
2010 
2011   parseChildBlock();
2012 }
2013 
2014 bool UnwrappedLineParser::tryToParseBracedList() {
2015   if (FormatTok->is(BK_Unknown))
2016     calculateBraceTypes();
2017   assert(FormatTok->isNot(BK_Unknown));
2018   if (FormatTok->is(BK_Block))
2019     return false;
2020   nextToken();
2021   parseBracedList();
2022   return true;
2023 }
2024 
2025 bool UnwrappedLineParser::tryToParseChildBlock() {
2026   assert(Style.isJavaScript() || Style.isCSharp());
2027   assert(FormatTok->is(TT_FatArrow));
2028   // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow.
2029   // They always start an expression or a child block if followed by a curly
2030   // brace.
2031   nextToken();
2032   if (FormatTok->isNot(tok::l_brace))
2033     return false;
2034   parseChildBlock();
2035   return true;
2036 }
2037 
2038 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons,
2039                                           bool IsEnum,
2040                                           tok::TokenKind ClosingBraceKind) {
2041   bool HasError = false;
2042 
2043   // FIXME: Once we have an expression parser in the UnwrappedLineParser,
2044   // replace this by using parseAssignmentExpression() inside.
2045   do {
2046     if (Style.isCSharp() && FormatTok->is(TT_FatArrow) &&
2047         tryToParseChildBlock())
2048       continue;
2049     if (Style.isJavaScript()) {
2050       if (FormatTok->is(Keywords.kw_function) ||
2051           FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) {
2052         tryToParseJSFunction();
2053         continue;
2054       }
2055       if (FormatTok->is(tok::l_brace)) {
2056         // Could be a method inside of a braced list `{a() { return 1; }}`.
2057         if (tryToParseBracedList())
2058           continue;
2059         parseChildBlock();
2060       }
2061     }
2062     if (FormatTok->Tok.getKind() == ClosingBraceKind) {
2063       if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
2064         addUnwrappedLine();
2065       nextToken();
2066       return !HasError;
2067     }
2068     switch (FormatTok->Tok.getKind()) {
2069     case tok::l_square:
2070       if (Style.isCSharp())
2071         parseSquare();
2072       else
2073         tryToParseLambda();
2074       break;
2075     case tok::l_paren:
2076       parseParens();
2077       // JavaScript can just have free standing methods and getters/setters in
2078       // object literals. Detect them by a "{" following ")".
2079       if (Style.isJavaScript()) {
2080         if (FormatTok->is(tok::l_brace))
2081           parseChildBlock();
2082         break;
2083       }
2084       break;
2085     case tok::l_brace:
2086       // Assume there are no blocks inside a braced init list apart
2087       // from the ones we explicitly parse out (like lambdas).
2088       FormatTok->setBlockKind(BK_BracedInit);
2089       nextToken();
2090       parseBracedList();
2091       break;
2092     case tok::less:
2093       if (Style.Language == FormatStyle::LK_Proto) {
2094         nextToken();
2095         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
2096                         /*ClosingBraceKind=*/tok::greater);
2097       } else {
2098         nextToken();
2099       }
2100       break;
2101     case tok::semi:
2102       // JavaScript (or more precisely TypeScript) can have semicolons in braced
2103       // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be
2104       // used for error recovery if we have otherwise determined that this is
2105       // a braced list.
2106       if (Style.isJavaScript()) {
2107         nextToken();
2108         break;
2109       }
2110       HasError = true;
2111       if (!ContinueOnSemicolons)
2112         return !HasError;
2113       nextToken();
2114       break;
2115     case tok::comma:
2116       nextToken();
2117       if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
2118         addUnwrappedLine();
2119       break;
2120     default:
2121       nextToken();
2122       break;
2123     }
2124   } while (!eof());
2125   return false;
2126 }
2127 
2128 void UnwrappedLineParser::parseParens() {
2129   assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected.");
2130   nextToken();
2131   do {
2132     switch (FormatTok->Tok.getKind()) {
2133     case tok::l_paren:
2134       parseParens();
2135       if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace))
2136         parseChildBlock();
2137       break;
2138     case tok::r_paren:
2139       nextToken();
2140       return;
2141     case tok::r_brace:
2142       // A "}" inside parenthesis is an error if there wasn't a matching "{".
2143       return;
2144     case tok::l_square:
2145       tryToParseLambda();
2146       break;
2147     case tok::l_brace:
2148       if (!tryToParseBracedList())
2149         parseChildBlock();
2150       break;
2151     case tok::at:
2152       nextToken();
2153       if (FormatTok->Tok.is(tok::l_brace)) {
2154         nextToken();
2155         parseBracedList();
2156       }
2157       break;
2158     case tok::equal:
2159       if (Style.isCSharp() && FormatTok->is(TT_FatArrow))
2160         tryToParseChildBlock();
2161       else
2162         nextToken();
2163       break;
2164     case tok::kw_class:
2165       if (Style.isJavaScript())
2166         parseRecord(/*ParseAsExpr=*/true);
2167       else
2168         nextToken();
2169       break;
2170     case tok::identifier:
2171       if (Style.isJavaScript() &&
2172           (FormatTok->is(Keywords.kw_function) ||
2173            FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)))
2174         tryToParseJSFunction();
2175       else
2176         nextToken();
2177       break;
2178     default:
2179       nextToken();
2180       break;
2181     }
2182   } while (!eof());
2183 }
2184 
2185 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) {
2186   if (!LambdaIntroducer) {
2187     assert(FormatTok->Tok.is(tok::l_square) && "'[' expected.");
2188     if (tryToParseLambda())
2189       return;
2190   }
2191   do {
2192     switch (FormatTok->Tok.getKind()) {
2193     case tok::l_paren:
2194       parseParens();
2195       break;
2196     case tok::r_square:
2197       nextToken();
2198       return;
2199     case tok::r_brace:
2200       // A "}" inside parenthesis is an error if there wasn't a matching "{".
2201       return;
2202     case tok::l_square:
2203       parseSquare();
2204       break;
2205     case tok::l_brace: {
2206       if (!tryToParseBracedList())
2207         parseChildBlock();
2208       break;
2209     }
2210     case tok::at:
2211       nextToken();
2212       if (FormatTok->Tok.is(tok::l_brace)) {
2213         nextToken();
2214         parseBracedList();
2215       }
2216       break;
2217     default:
2218       nextToken();
2219       break;
2220     }
2221   } while (!eof());
2222 }
2223 
2224 void UnwrappedLineParser::keepAncestorBraces() {
2225   if (!Style.RemoveBracesLLVM)
2226     return;
2227 
2228   const int MaxNestingLevels = 2;
2229   const int Size = NestedTooDeep.size();
2230   if (Size >= MaxNestingLevels)
2231     NestedTooDeep[Size - MaxNestingLevels] = true;
2232   NestedTooDeep.push_back(false);
2233 }
2234 
2235 static void markOptionalBraces(FormatToken *LeftBrace) {
2236   if (!LeftBrace)
2237     return;
2238 
2239   assert(LeftBrace->is(tok::l_brace));
2240 
2241   FormatToken *RightBrace = LeftBrace->MatchingParen;
2242   if (!RightBrace) {
2243     assert(!LeftBrace->Optional);
2244     return;
2245   }
2246 
2247   assert(RightBrace->is(tok::r_brace));
2248   assert(RightBrace->MatchingParen == LeftBrace);
2249   assert(LeftBrace->Optional == RightBrace->Optional);
2250 
2251   LeftBrace->Optional = true;
2252   RightBrace->Optional = true;
2253 }
2254 
2255 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind,
2256                                                   bool KeepBraces) {
2257   auto HandleAttributes = [this]() {
2258     // Handle AttributeMacro, e.g. `if (x) UNLIKELY`.
2259     if (FormatTok->is(TT_AttributeMacro))
2260       nextToken();
2261     // Handle [[likely]] / [[unlikely]] attributes.
2262     if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute())
2263       parseSquare();
2264   };
2265 
2266   assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected");
2267   nextToken();
2268   if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier))
2269     nextToken();
2270   if (FormatTok->Tok.is(tok::l_paren))
2271     parseParens();
2272   HandleAttributes();
2273 
2274   bool NeedsUnwrappedLine = false;
2275   keepAncestorBraces();
2276 
2277   FormatToken *IfLeftBrace = nullptr;
2278   IfStmtKind IfBlockKind = IfStmtKind::NotIf;
2279 
2280   if (FormatTok->Tok.is(tok::l_brace)) {
2281     IfLeftBrace = FormatTok;
2282     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2283     IfBlockKind = parseBlock();
2284     if (Style.BraceWrapping.BeforeElse)
2285       addUnwrappedLine();
2286     else
2287       NeedsUnwrappedLine = true;
2288   } else {
2289     addUnwrappedLine();
2290     ++Line->Level;
2291     parseStructuralElement();
2292     --Line->Level;
2293   }
2294 
2295   bool KeepIfBraces = false;
2296   if (Style.RemoveBracesLLVM) {
2297     assert(!NestedTooDeep.empty());
2298     KeepIfBraces = (IfLeftBrace && !IfLeftBrace->MatchingParen) ||
2299                    NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly ||
2300                    IfBlockKind == IfStmtKind::IfElseIf;
2301   }
2302 
2303   FormatToken *ElseLeftBrace = nullptr;
2304   IfStmtKind Kind = IfStmtKind::IfOnly;
2305 
2306   if (FormatTok->Tok.is(tok::kw_else)) {
2307     if (Style.RemoveBracesLLVM) {
2308       NestedTooDeep.back() = false;
2309       Kind = IfStmtKind::IfElse;
2310     }
2311     nextToken();
2312     HandleAttributes();
2313     if (FormatTok->Tok.is(tok::l_brace)) {
2314       ElseLeftBrace = FormatTok;
2315       CompoundStatementIndenter Indenter(this, Style, Line->Level);
2316       if (parseBlock() == IfStmtKind::IfOnly)
2317         Kind = IfStmtKind::IfElseIf;
2318       addUnwrappedLine();
2319     } else if (FormatTok->Tok.is(tok::kw_if)) {
2320       FormatToken *Previous = Tokens->getPreviousToken();
2321       const bool IsPrecededByComment = Previous && Previous->is(tok::comment);
2322       if (IsPrecededByComment) {
2323         addUnwrappedLine();
2324         ++Line->Level;
2325       }
2326       bool TooDeep = true;
2327       if (Style.RemoveBracesLLVM) {
2328         Kind = IfStmtKind::IfElseIf;
2329         TooDeep = NestedTooDeep.pop_back_val();
2330       }
2331       ElseLeftBrace =
2332           parseIfThenElse(/*IfKind=*/nullptr, KeepBraces || KeepIfBraces);
2333       if (Style.RemoveBracesLLVM)
2334         NestedTooDeep.push_back(TooDeep);
2335       if (IsPrecededByComment)
2336         --Line->Level;
2337     } else {
2338       addUnwrappedLine();
2339       ++Line->Level;
2340       parseStructuralElement();
2341       if (FormatTok->is(tok::eof))
2342         addUnwrappedLine();
2343       --Line->Level;
2344     }
2345   } else {
2346     if (Style.RemoveBracesLLVM)
2347       KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse;
2348     if (NeedsUnwrappedLine)
2349       addUnwrappedLine();
2350   }
2351 
2352   if (!Style.RemoveBracesLLVM)
2353     return nullptr;
2354 
2355   assert(!NestedTooDeep.empty());
2356   const bool KeepElseBraces =
2357       (ElseLeftBrace && !ElseLeftBrace->MatchingParen) || NestedTooDeep.back();
2358 
2359   NestedTooDeep.pop_back();
2360 
2361   if (!KeepBraces && !KeepIfBraces && !KeepElseBraces) {
2362     markOptionalBraces(IfLeftBrace);
2363     markOptionalBraces(ElseLeftBrace);
2364   } else if (IfLeftBrace) {
2365     FormatToken *IfRightBrace = IfLeftBrace->MatchingParen;
2366     if (IfRightBrace) {
2367       assert(IfRightBrace->MatchingParen == IfLeftBrace);
2368       assert(!IfLeftBrace->Optional);
2369       assert(!IfRightBrace->Optional);
2370       IfLeftBrace->MatchingParen = nullptr;
2371       IfRightBrace->MatchingParen = nullptr;
2372     }
2373   }
2374 
2375   if (IfKind)
2376     *IfKind = Kind;
2377 
2378   return IfLeftBrace;
2379 }
2380 
2381 void UnwrappedLineParser::parseTryCatch() {
2382   assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
2383   nextToken();
2384   bool NeedsUnwrappedLine = false;
2385   if (FormatTok->is(tok::colon)) {
2386     // We are in a function try block, what comes is an initializer list.
2387     nextToken();
2388 
2389     // In case identifiers were removed by clang-tidy, what might follow is
2390     // multiple commas in sequence - before the first identifier.
2391     while (FormatTok->is(tok::comma))
2392       nextToken();
2393 
2394     while (FormatTok->is(tok::identifier)) {
2395       nextToken();
2396       if (FormatTok->is(tok::l_paren))
2397         parseParens();
2398       if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) &&
2399           FormatTok->is(tok::l_brace)) {
2400         do {
2401           nextToken();
2402         } while (!FormatTok->is(tok::r_brace));
2403         nextToken();
2404       }
2405 
2406       // In case identifiers were removed by clang-tidy, what might follow is
2407       // multiple commas in sequence - after the first identifier.
2408       while (FormatTok->is(tok::comma))
2409         nextToken();
2410     }
2411   }
2412   // Parse try with resource.
2413   if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) {
2414     parseParens();
2415   }
2416 
2417   keepAncestorBraces();
2418 
2419   if (FormatTok->is(tok::l_brace)) {
2420     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2421     parseBlock();
2422     if (Style.BraceWrapping.BeforeCatch) {
2423       addUnwrappedLine();
2424     } else {
2425       NeedsUnwrappedLine = true;
2426     }
2427   } else if (!FormatTok->is(tok::kw_catch)) {
2428     // The C++ standard requires a compound-statement after a try.
2429     // If there's none, we try to assume there's a structuralElement
2430     // and try to continue.
2431     addUnwrappedLine();
2432     ++Line->Level;
2433     parseStructuralElement();
2434     --Line->Level;
2435   }
2436   while (true) {
2437     if (FormatTok->is(tok::at))
2438       nextToken();
2439     if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except,
2440                              tok::kw___finally) ||
2441           ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) &&
2442            FormatTok->is(Keywords.kw_finally)) ||
2443           (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) ||
2444            FormatTok->Tok.isObjCAtKeyword(tok::objc_finally))))
2445       break;
2446     nextToken();
2447     while (FormatTok->isNot(tok::l_brace)) {
2448       if (FormatTok->is(tok::l_paren)) {
2449         parseParens();
2450         continue;
2451       }
2452       if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) {
2453         if (Style.RemoveBracesLLVM)
2454           NestedTooDeep.pop_back();
2455         return;
2456       }
2457       nextToken();
2458     }
2459     NeedsUnwrappedLine = false;
2460     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2461     parseBlock();
2462     if (Style.BraceWrapping.BeforeCatch)
2463       addUnwrappedLine();
2464     else
2465       NeedsUnwrappedLine = true;
2466   }
2467 
2468   if (Style.RemoveBracesLLVM)
2469     NestedTooDeep.pop_back();
2470 
2471   if (NeedsUnwrappedLine)
2472     addUnwrappedLine();
2473 }
2474 
2475 void UnwrappedLineParser::parseNamespace() {
2476   assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) &&
2477          "'namespace' expected");
2478 
2479   const FormatToken &InitialToken = *FormatTok;
2480   nextToken();
2481   if (InitialToken.is(TT_NamespaceMacro)) {
2482     parseParens();
2483   } else {
2484     while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline,
2485                               tok::l_square, tok::period) ||
2486            (Style.isCSharp() && FormatTok->is(tok::kw_union))) {
2487       if (FormatTok->is(tok::l_square))
2488         parseSquare();
2489       else
2490         nextToken();
2491     }
2492   }
2493   if (FormatTok->Tok.is(tok::l_brace)) {
2494     if (ShouldBreakBeforeBrace(Style, InitialToken))
2495       addUnwrappedLine();
2496 
2497     unsigned AddLevels =
2498         Style.NamespaceIndentation == FormatStyle::NI_All ||
2499                 (Style.NamespaceIndentation == FormatStyle::NI_Inner &&
2500                  DeclarationScopeStack.size() > 1)
2501             ? 1u
2502             : 0u;
2503     bool ManageWhitesmithsBraces =
2504         AddLevels == 0u &&
2505         Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
2506 
2507     // If we're in Whitesmiths mode, indent the brace if we're not indenting
2508     // the whole block.
2509     if (ManageWhitesmithsBraces)
2510       ++Line->Level;
2511 
2512     parseBlock(/*MustBeDeclaration=*/true, AddLevels,
2513                /*MunchSemi=*/true,
2514                /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces);
2515 
2516     // Munch the semicolon after a namespace. This is more common than one would
2517     // think. Putting the semicolon into its own line is very ugly.
2518     if (FormatTok->Tok.is(tok::semi))
2519       nextToken();
2520 
2521     addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep);
2522 
2523     if (ManageWhitesmithsBraces)
2524       --Line->Level;
2525   }
2526   // FIXME: Add error handling.
2527 }
2528 
2529 void UnwrappedLineParser::parseNew() {
2530   assert(FormatTok->is(tok::kw_new) && "'new' expected");
2531   nextToken();
2532 
2533   if (Style.isCSharp()) {
2534     do {
2535       if (FormatTok->is(tok::l_brace))
2536         parseBracedList();
2537 
2538       if (FormatTok->isOneOf(tok::semi, tok::comma))
2539         return;
2540 
2541       nextToken();
2542     } while (!eof());
2543   }
2544 
2545   if (Style.Language != FormatStyle::LK_Java)
2546     return;
2547 
2548   // In Java, we can parse everything up to the parens, which aren't optional.
2549   do {
2550     // There should not be a ;, { or } before the new's open paren.
2551     if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace))
2552       return;
2553 
2554     // Consume the parens.
2555     if (FormatTok->is(tok::l_paren)) {
2556       parseParens();
2557 
2558       // If there is a class body of an anonymous class, consume that as child.
2559       if (FormatTok->is(tok::l_brace))
2560         parseChildBlock();
2561       return;
2562     }
2563     nextToken();
2564   } while (!eof());
2565 }
2566 
2567 void UnwrappedLineParser::parseForOrWhileLoop() {
2568   assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) &&
2569          "'for', 'while' or foreach macro expected");
2570   nextToken();
2571   // JS' for await ( ...
2572   if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await))
2573     nextToken();
2574   if (Style.isCpp() && FormatTok->is(tok::kw_co_await))
2575     nextToken();
2576   if (FormatTok->Tok.is(tok::l_paren))
2577     parseParens();
2578 
2579   keepAncestorBraces();
2580 
2581   if (FormatTok->Tok.is(tok::l_brace)) {
2582     FormatToken *LeftBrace = FormatTok;
2583     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2584     parseBlock();
2585     if (Style.RemoveBracesLLVM) {
2586       assert(!NestedTooDeep.empty());
2587       if (!NestedTooDeep.back())
2588         markOptionalBraces(LeftBrace);
2589     }
2590     addUnwrappedLine();
2591   } else {
2592     addUnwrappedLine();
2593     ++Line->Level;
2594     parseStructuralElement();
2595     --Line->Level;
2596   }
2597 
2598   if (Style.RemoveBracesLLVM)
2599     NestedTooDeep.pop_back();
2600 }
2601 
2602 void UnwrappedLineParser::parseDoWhile() {
2603   assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected");
2604   nextToken();
2605 
2606   keepAncestorBraces();
2607 
2608   if (FormatTok->Tok.is(tok::l_brace)) {
2609     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2610     parseBlock();
2611     if (Style.BraceWrapping.BeforeWhile)
2612       addUnwrappedLine();
2613   } else {
2614     addUnwrappedLine();
2615     ++Line->Level;
2616     parseStructuralElement();
2617     --Line->Level;
2618   }
2619 
2620   if (Style.RemoveBracesLLVM)
2621     NestedTooDeep.pop_back();
2622 
2623   // FIXME: Add error handling.
2624   if (!FormatTok->Tok.is(tok::kw_while)) {
2625     addUnwrappedLine();
2626     return;
2627   }
2628 
2629   // If in Whitesmiths mode, the line with the while() needs to be indented
2630   // to the same level as the block.
2631   if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
2632     ++Line->Level;
2633 
2634   nextToken();
2635   parseStructuralElement();
2636 }
2637 
2638 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) {
2639   nextToken();
2640   unsigned OldLineLevel = Line->Level;
2641   if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0))
2642     --Line->Level;
2643   if (LeftAlignLabel)
2644     Line->Level = 0;
2645 
2646   if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() &&
2647       FormatTok->Tok.is(tok::l_brace)) {
2648 
2649     CompoundStatementIndenter Indenter(this, Line->Level,
2650                                        Style.BraceWrapping.AfterCaseLabel,
2651                                        Style.BraceWrapping.IndentBraces);
2652     parseBlock();
2653     if (FormatTok->Tok.is(tok::kw_break)) {
2654       if (Style.BraceWrapping.AfterControlStatement ==
2655           FormatStyle::BWACS_Always) {
2656         addUnwrappedLine();
2657         if (!Style.IndentCaseBlocks &&
2658             Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
2659           ++Line->Level;
2660         }
2661       }
2662       parseStructuralElement();
2663     }
2664     addUnwrappedLine();
2665   } else {
2666     if (FormatTok->is(tok::semi))
2667       nextToken();
2668     addUnwrappedLine();
2669   }
2670   Line->Level = OldLineLevel;
2671   if (FormatTok->isNot(tok::l_brace)) {
2672     parseStructuralElement();
2673     addUnwrappedLine();
2674   }
2675 }
2676 
2677 void UnwrappedLineParser::parseCaseLabel() {
2678   assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected");
2679 
2680   // FIXME: fix handling of complex expressions here.
2681   do {
2682     nextToken();
2683   } while (!eof() && !FormatTok->Tok.is(tok::colon));
2684   parseLabel();
2685 }
2686 
2687 void UnwrappedLineParser::parseSwitch() {
2688   assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected");
2689   nextToken();
2690   if (FormatTok->Tok.is(tok::l_paren))
2691     parseParens();
2692 
2693   keepAncestorBraces();
2694 
2695   if (FormatTok->Tok.is(tok::l_brace)) {
2696     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2697     parseBlock();
2698     addUnwrappedLine();
2699   } else {
2700     addUnwrappedLine();
2701     ++Line->Level;
2702     parseStructuralElement();
2703     --Line->Level;
2704   }
2705 
2706   if (Style.RemoveBracesLLVM)
2707     NestedTooDeep.pop_back();
2708 }
2709 
2710 void UnwrappedLineParser::parseAccessSpecifier() {
2711   FormatToken *AccessSpecifierCandidate = FormatTok;
2712   nextToken();
2713   // Understand Qt's slots.
2714   if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots))
2715     nextToken();
2716   // Otherwise, we don't know what it is, and we'd better keep the next token.
2717   if (FormatTok->Tok.is(tok::colon)) {
2718     nextToken();
2719     addUnwrappedLine();
2720   } else if (!FormatTok->Tok.is(tok::coloncolon) &&
2721              !std::binary_search(COperatorsFollowingVar.begin(),
2722                                  COperatorsFollowingVar.end(),
2723                                  FormatTok->Tok.getKind())) {
2724     // Not a variable name nor namespace name.
2725     addUnwrappedLine();
2726   } else if (AccessSpecifierCandidate) {
2727     // Consider the access specifier to be a C identifier.
2728     AccessSpecifierCandidate->Tok.setKind(tok::identifier);
2729   }
2730 }
2731 
2732 void UnwrappedLineParser::parseConcept() {
2733   assert(FormatTok->Tok.is(tok::kw_concept) && "'concept' expected");
2734   nextToken();
2735   if (!FormatTok->Tok.is(tok::identifier))
2736     return;
2737   nextToken();
2738   if (!FormatTok->Tok.is(tok::equal))
2739     return;
2740   nextToken();
2741   if (FormatTok->Tok.is(tok::kw_requires)) {
2742     nextToken();
2743     parseRequiresExpression(Line->Level);
2744   } else {
2745     parseConstraintExpression(Line->Level);
2746   }
2747 }
2748 
2749 void UnwrappedLineParser::parseRequiresExpression(unsigned int OriginalLevel) {
2750   // requires (R range)
2751   if (FormatTok->Tok.is(tok::l_paren)) {
2752     parseParens();
2753     if (Style.IndentRequires && OriginalLevel != Line->Level) {
2754       addUnwrappedLine();
2755       --Line->Level;
2756     }
2757   }
2758 
2759   if (FormatTok->Tok.is(tok::l_brace)) {
2760     if (Style.BraceWrapping.AfterFunction)
2761       addUnwrappedLine();
2762     FormatTok->setType(TT_FunctionLBrace);
2763     parseBlock();
2764     addUnwrappedLine();
2765   } else {
2766     parseConstraintExpression(OriginalLevel);
2767   }
2768 }
2769 
2770 void UnwrappedLineParser::parseConstraintExpression(
2771     unsigned int OriginalLevel) {
2772   // requires Id<T> && Id<T> || Id<T>
2773   while (
2774       FormatTok->isOneOf(tok::identifier, tok::kw_requires, tok::coloncolon)) {
2775     nextToken();
2776     while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::less,
2777                               tok::greater, tok::comma, tok::ellipsis)) {
2778       if (FormatTok->Tok.is(tok::less)) {
2779         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
2780                         /*ClosingBraceKind=*/tok::greater);
2781         continue;
2782       }
2783       nextToken();
2784     }
2785     if (FormatTok->Tok.is(tok::kw_requires)) {
2786       parseRequiresExpression(OriginalLevel);
2787     }
2788     if (FormatTok->Tok.is(tok::less)) {
2789       parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
2790                       /*ClosingBraceKind=*/tok::greater);
2791     }
2792 
2793     if (FormatTok->Tok.is(tok::l_paren)) {
2794       parseParens();
2795     }
2796     if (FormatTok->Tok.is(tok::l_brace)) {
2797       if (Style.BraceWrapping.AfterFunction)
2798         addUnwrappedLine();
2799       FormatTok->setType(TT_FunctionLBrace);
2800       parseBlock();
2801     }
2802     if (FormatTok->Tok.is(tok::semi)) {
2803       // Eat any trailing semi.
2804       nextToken();
2805       addUnwrappedLine();
2806     }
2807     if (FormatTok->Tok.is(tok::colon)) {
2808       return;
2809     }
2810     if (!FormatTok->Tok.isOneOf(tok::ampamp, tok::pipepipe)) {
2811       if (FormatTok->Previous &&
2812           !FormatTok->Previous->isOneOf(tok::identifier, tok::kw_requires,
2813                                         tok::coloncolon)) {
2814         addUnwrappedLine();
2815       }
2816       if (Style.IndentRequires && OriginalLevel != Line->Level) {
2817         --Line->Level;
2818       }
2819       break;
2820     } else {
2821       FormatTok->setType(TT_ConstraintJunctions);
2822     }
2823 
2824     nextToken();
2825   }
2826 }
2827 
2828 void UnwrappedLineParser::parseRequires() {
2829   assert(FormatTok->Tok.is(tok::kw_requires) && "'requires' expected");
2830 
2831   unsigned OriginalLevel = Line->Level;
2832   if (FormatTok->Previous && FormatTok->Previous->is(tok::greater)) {
2833     addUnwrappedLine();
2834     if (Style.IndentRequires) {
2835       ++Line->Level;
2836     }
2837   }
2838   nextToken();
2839 
2840   parseRequiresExpression(OriginalLevel);
2841 }
2842 
2843 bool UnwrappedLineParser::parseEnum() {
2844   const FormatToken &InitialToken = *FormatTok;
2845 
2846   // Won't be 'enum' for NS_ENUMs.
2847   if (FormatTok->Tok.is(tok::kw_enum))
2848     nextToken();
2849 
2850   // In TypeScript, "enum" can also be used as property name, e.g. in interface
2851   // declarations. An "enum" keyword followed by a colon would be a syntax
2852   // error and thus assume it is just an identifier.
2853   if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question))
2854     return false;
2855 
2856   // In protobuf, "enum" can be used as a field name.
2857   if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal))
2858     return false;
2859 
2860   // Eat up enum class ...
2861   if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct))
2862     nextToken();
2863 
2864   while (FormatTok->Tok.getIdentifierInfo() ||
2865          FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less,
2866                             tok::greater, tok::comma, tok::question)) {
2867     nextToken();
2868     // We can have macros or attributes in between 'enum' and the enum name.
2869     if (FormatTok->is(tok::l_paren))
2870       parseParens();
2871     if (FormatTok->is(tok::identifier)) {
2872       nextToken();
2873       // If there are two identifiers in a row, this is likely an elaborate
2874       // return type. In Java, this can be "implements", etc.
2875       if (Style.isCpp() && FormatTok->is(tok::identifier))
2876         return false;
2877     }
2878   }
2879 
2880   // Just a declaration or something is wrong.
2881   if (FormatTok->isNot(tok::l_brace))
2882     return true;
2883   FormatTok->setType(TT_RecordLBrace);
2884   FormatTok->setBlockKind(BK_Block);
2885 
2886   if (Style.Language == FormatStyle::LK_Java) {
2887     // Java enums are different.
2888     parseJavaEnumBody();
2889     return true;
2890   }
2891   if (Style.Language == FormatStyle::LK_Proto) {
2892     parseBlock(/*MustBeDeclaration=*/true);
2893     return true;
2894   }
2895 
2896   if (!Style.AllowShortEnumsOnASingleLine &&
2897       ShouldBreakBeforeBrace(Style, InitialToken))
2898     addUnwrappedLine();
2899   // Parse enum body.
2900   nextToken();
2901   if (!Style.AllowShortEnumsOnASingleLine) {
2902     addUnwrappedLine();
2903     Line->Level += 1;
2904   }
2905   bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true,
2906                                    /*IsEnum=*/true);
2907   if (!Style.AllowShortEnumsOnASingleLine)
2908     Line->Level -= 1;
2909   if (HasError) {
2910     if (FormatTok->is(tok::semi))
2911       nextToken();
2912     addUnwrappedLine();
2913   }
2914   return true;
2915 
2916   // There is no addUnwrappedLine() here so that we fall through to parsing a
2917   // structural element afterwards. Thus, in "enum A {} n, m;",
2918   // "} n, m;" will end up in one unwrapped line.
2919 }
2920 
2921 bool UnwrappedLineParser::parseStructLike() {
2922   // parseRecord falls through and does not yet add an unwrapped line as a
2923   // record declaration or definition can start a structural element.
2924   parseRecord();
2925   // This does not apply to Java, JavaScript and C#.
2926   if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
2927       Style.isCSharp()) {
2928     if (FormatTok->is(tok::semi))
2929       nextToken();
2930     addUnwrappedLine();
2931     return true;
2932   }
2933   return false;
2934 }
2935 
2936 namespace {
2937 // A class used to set and restore the Token position when peeking
2938 // ahead in the token source.
2939 class ScopedTokenPosition {
2940   unsigned StoredPosition;
2941   FormatTokenSource *Tokens;
2942 
2943 public:
2944   ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) {
2945     assert(Tokens && "Tokens expected to not be null");
2946     StoredPosition = Tokens->getPosition();
2947   }
2948 
2949   ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); }
2950 };
2951 } // namespace
2952 
2953 // Look to see if we have [[ by looking ahead, if
2954 // its not then rewind to the original position.
2955 bool UnwrappedLineParser::tryToParseSimpleAttribute() {
2956   ScopedTokenPosition AutoPosition(Tokens);
2957   FormatToken *Tok = Tokens->getNextToken();
2958   // We already read the first [ check for the second.
2959   if (!Tok->is(tok::l_square)) {
2960     return false;
2961   }
2962   // Double check that the attribute is just something
2963   // fairly simple.
2964   while (Tok->isNot(tok::eof)) {
2965     if (Tok->is(tok::r_square)) {
2966       break;
2967     }
2968     Tok = Tokens->getNextToken();
2969   }
2970   if (Tok->is(tok::eof))
2971     return false;
2972   Tok = Tokens->getNextToken();
2973   if (!Tok->is(tok::r_square)) {
2974     return false;
2975   }
2976   Tok = Tokens->getNextToken();
2977   if (Tok->is(tok::semi)) {
2978     return false;
2979   }
2980   return true;
2981 }
2982 
2983 void UnwrappedLineParser::parseJavaEnumBody() {
2984   // Determine whether the enum is simple, i.e. does not have a semicolon or
2985   // constants with class bodies. Simple enums can be formatted like braced
2986   // lists, contracted to a single line, etc.
2987   unsigned StoredPosition = Tokens->getPosition();
2988   bool IsSimple = true;
2989   FormatToken *Tok = Tokens->getNextToken();
2990   while (!Tok->is(tok::eof)) {
2991     if (Tok->is(tok::r_brace))
2992       break;
2993     if (Tok->isOneOf(tok::l_brace, tok::semi)) {
2994       IsSimple = false;
2995       break;
2996     }
2997     // FIXME: This will also mark enums with braces in the arguments to enum
2998     // constants as "not simple". This is probably fine in practice, though.
2999     Tok = Tokens->getNextToken();
3000   }
3001   FormatTok = Tokens->setPosition(StoredPosition);
3002 
3003   if (IsSimple) {
3004     nextToken();
3005     parseBracedList();
3006     addUnwrappedLine();
3007     return;
3008   }
3009 
3010   // Parse the body of a more complex enum.
3011   // First add a line for everything up to the "{".
3012   nextToken();
3013   addUnwrappedLine();
3014   ++Line->Level;
3015 
3016   // Parse the enum constants.
3017   while (FormatTok) {
3018     if (FormatTok->is(tok::l_brace)) {
3019       // Parse the constant's class body.
3020       parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u,
3021                  /*MunchSemi=*/false);
3022     } else if (FormatTok->is(tok::l_paren)) {
3023       parseParens();
3024     } else if (FormatTok->is(tok::comma)) {
3025       nextToken();
3026       addUnwrappedLine();
3027     } else if (FormatTok->is(tok::semi)) {
3028       nextToken();
3029       addUnwrappedLine();
3030       break;
3031     } else if (FormatTok->is(tok::r_brace)) {
3032       addUnwrappedLine();
3033       break;
3034     } else {
3035       nextToken();
3036     }
3037   }
3038 
3039   // Parse the class body after the enum's ";" if any.
3040   parseLevel(/*HasOpeningBrace=*/true);
3041   nextToken();
3042   --Line->Level;
3043   addUnwrappedLine();
3044 }
3045 
3046 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) {
3047   const FormatToken &InitialToken = *FormatTok;
3048   nextToken();
3049 
3050   // The actual identifier can be a nested name specifier, and in macros
3051   // it is often token-pasted.
3052   // An [[attribute]] can be before the identifier.
3053   while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash,
3054                             tok::kw___attribute, tok::kw___declspec,
3055                             tok::kw_alignas, tok::l_square, tok::r_square) ||
3056          ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) &&
3057           FormatTok->isOneOf(tok::period, tok::comma))) {
3058     if (Style.isJavaScript() &&
3059         FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) {
3060       // JavaScript/TypeScript supports inline object types in
3061       // extends/implements positions:
3062       //     class Foo implements {bar: number} { }
3063       nextToken();
3064       if (FormatTok->is(tok::l_brace)) {
3065         tryToParseBracedList();
3066         continue;
3067       }
3068     }
3069     bool IsNonMacroIdentifier =
3070         FormatTok->is(tok::identifier) &&
3071         FormatTok->TokenText != FormatTok->TokenText.upper();
3072     nextToken();
3073     // We can have macros or attributes in between 'class' and the class name.
3074     if (!IsNonMacroIdentifier) {
3075       if (FormatTok->Tok.is(tok::l_paren)) {
3076         parseParens();
3077       } else if (FormatTok->is(TT_AttributeSquare)) {
3078         parseSquare();
3079         // Consume the closing TT_AttributeSquare.
3080         if (FormatTok->Next && FormatTok->is(TT_AttributeSquare))
3081           nextToken();
3082       }
3083     }
3084   }
3085 
3086   // Note that parsing away template declarations here leads to incorrectly
3087   // accepting function declarations as record declarations.
3088   // In general, we cannot solve this problem. Consider:
3089   // class A<int> B() {}
3090   // which can be a function definition or a class definition when B() is a
3091   // macro. If we find enough real-world cases where this is a problem, we
3092   // can parse for the 'template' keyword in the beginning of the statement,
3093   // and thus rule out the record production in case there is no template
3094   // (this would still leave us with an ambiguity between template function
3095   // and class declarations).
3096   if (FormatTok->isOneOf(tok::colon, tok::less)) {
3097     while (!eof()) {
3098       if (FormatTok->is(tok::l_brace)) {
3099         calculateBraceTypes(/*ExpectClassBody=*/true);
3100         if (!tryToParseBracedList())
3101           break;
3102       }
3103       if (FormatTok->is(tok::l_square)) {
3104         FormatToken *Previous = FormatTok->Previous;
3105         if (!Previous ||
3106             !(Previous->is(tok::r_paren) || Previous->isTypeOrIdentifier())) {
3107           // Don't try parsing a lambda if we had a closing parenthesis before,
3108           // it was probably a pointer to an array: int (*)[].
3109           if (!tryToParseLambda())
3110             break;
3111         }
3112       }
3113       if (FormatTok->Tok.is(tok::semi))
3114         return;
3115       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) {
3116         addUnwrappedLine();
3117         nextToken();
3118         parseCSharpGenericTypeConstraint();
3119         break;
3120       }
3121       nextToken();
3122     }
3123   }
3124   if (FormatTok->Tok.is(tok::l_brace)) {
3125     FormatTok->setType(TT_RecordLBrace);
3126     if (ParseAsExpr) {
3127       parseChildBlock();
3128     } else {
3129       if (ShouldBreakBeforeBrace(Style, InitialToken))
3130         addUnwrappedLine();
3131 
3132       unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u;
3133       parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false);
3134     }
3135   }
3136   // There is no addUnwrappedLine() here so that we fall through to parsing a
3137   // structural element afterwards. Thus, in "class A {} n, m;",
3138   // "} n, m;" will end up in one unwrapped line.
3139 }
3140 
3141 void UnwrappedLineParser::parseObjCMethod() {
3142   assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) &&
3143          "'(' or identifier expected.");
3144   do {
3145     if (FormatTok->Tok.is(tok::semi)) {
3146       nextToken();
3147       addUnwrappedLine();
3148       return;
3149     } else if (FormatTok->Tok.is(tok::l_brace)) {
3150       if (Style.BraceWrapping.AfterFunction)
3151         addUnwrappedLine();
3152       parseBlock();
3153       addUnwrappedLine();
3154       return;
3155     } else {
3156       nextToken();
3157     }
3158   } while (!eof());
3159 }
3160 
3161 void UnwrappedLineParser::parseObjCProtocolList() {
3162   assert(FormatTok->Tok.is(tok::less) && "'<' expected.");
3163   do {
3164     nextToken();
3165     // Early exit in case someone forgot a close angle.
3166     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
3167         FormatTok->Tok.isObjCAtKeyword(tok::objc_end))
3168       return;
3169   } while (!eof() && FormatTok->Tok.isNot(tok::greater));
3170   nextToken(); // Skip '>'.
3171 }
3172 
3173 void UnwrappedLineParser::parseObjCUntilAtEnd() {
3174   do {
3175     if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) {
3176       nextToken();
3177       addUnwrappedLine();
3178       break;
3179     }
3180     if (FormatTok->is(tok::l_brace)) {
3181       parseBlock();
3182       // In ObjC interfaces, nothing should be following the "}".
3183       addUnwrappedLine();
3184     } else if (FormatTok->is(tok::r_brace)) {
3185       // Ignore stray "}". parseStructuralElement doesn't consume them.
3186       nextToken();
3187       addUnwrappedLine();
3188     } else if (FormatTok->isOneOf(tok::minus, tok::plus)) {
3189       nextToken();
3190       parseObjCMethod();
3191     } else {
3192       parseStructuralElement();
3193     }
3194   } while (!eof());
3195 }
3196 
3197 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() {
3198   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface ||
3199          FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation);
3200   nextToken();
3201   nextToken(); // interface name
3202 
3203   // @interface can be followed by a lightweight generic
3204   // specialization list, then either a base class or a category.
3205   if (FormatTok->Tok.is(tok::less)) {
3206     parseObjCLightweightGenerics();
3207   }
3208   if (FormatTok->Tok.is(tok::colon)) {
3209     nextToken();
3210     nextToken(); // base class name
3211     // The base class can also have lightweight generics applied to it.
3212     if (FormatTok->Tok.is(tok::less)) {
3213       parseObjCLightweightGenerics();
3214     }
3215   } else if (FormatTok->Tok.is(tok::l_paren))
3216     // Skip category, if present.
3217     parseParens();
3218 
3219   if (FormatTok->Tok.is(tok::less))
3220     parseObjCProtocolList();
3221 
3222   if (FormatTok->Tok.is(tok::l_brace)) {
3223     if (Style.BraceWrapping.AfterObjCDeclaration)
3224       addUnwrappedLine();
3225     parseBlock(/*MustBeDeclaration=*/true);
3226   }
3227 
3228   // With instance variables, this puts '}' on its own line.  Without instance
3229   // variables, this ends the @interface line.
3230   addUnwrappedLine();
3231 
3232   parseObjCUntilAtEnd();
3233 }
3234 
3235 void UnwrappedLineParser::parseObjCLightweightGenerics() {
3236   assert(FormatTok->Tok.is(tok::less));
3237   // Unlike protocol lists, generic parameterizations support
3238   // nested angles:
3239   //
3240   // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> :
3241   //     NSObject <NSCopying, NSSecureCoding>
3242   //
3243   // so we need to count how many open angles we have left.
3244   unsigned NumOpenAngles = 1;
3245   do {
3246     nextToken();
3247     // Early exit in case someone forgot a close angle.
3248     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
3249         FormatTok->Tok.isObjCAtKeyword(tok::objc_end))
3250       break;
3251     if (FormatTok->Tok.is(tok::less))
3252       ++NumOpenAngles;
3253     else if (FormatTok->Tok.is(tok::greater)) {
3254       assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative");
3255       --NumOpenAngles;
3256     }
3257   } while (!eof() && NumOpenAngles != 0);
3258   nextToken(); // Skip '>'.
3259 }
3260 
3261 // Returns true for the declaration/definition form of @protocol,
3262 // false for the expression form.
3263 bool UnwrappedLineParser::parseObjCProtocol() {
3264   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol);
3265   nextToken();
3266 
3267   if (FormatTok->is(tok::l_paren))
3268     // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);".
3269     return false;
3270 
3271   // The definition/declaration form,
3272   // @protocol Foo
3273   // - (int)someMethod;
3274   // @end
3275 
3276   nextToken(); // protocol name
3277 
3278   if (FormatTok->Tok.is(tok::less))
3279     parseObjCProtocolList();
3280 
3281   // Check for protocol declaration.
3282   if (FormatTok->Tok.is(tok::semi)) {
3283     nextToken();
3284     addUnwrappedLine();
3285     return true;
3286   }
3287 
3288   addUnwrappedLine();
3289   parseObjCUntilAtEnd();
3290   return true;
3291 }
3292 
3293 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
3294   bool IsImport = FormatTok->is(Keywords.kw_import);
3295   assert(IsImport || FormatTok->is(tok::kw_export));
3296   nextToken();
3297 
3298   // Consume the "default" in "export default class/function".
3299   if (FormatTok->is(tok::kw_default))
3300     nextToken();
3301 
3302   // Consume "async function", "function" and "default function", so that these
3303   // get parsed as free-standing JS functions, i.e. do not require a trailing
3304   // semicolon.
3305   if (FormatTok->is(Keywords.kw_async))
3306     nextToken();
3307   if (FormatTok->is(Keywords.kw_function)) {
3308     nextToken();
3309     return;
3310   }
3311 
3312   // For imports, `export *`, `export {...}`, consume the rest of the line up
3313   // to the terminating `;`. For everything else, just return and continue
3314   // parsing the structural element, i.e. the declaration or expression for
3315   // `export default`.
3316   if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) &&
3317       !FormatTok->isStringLiteral())
3318     return;
3319 
3320   while (!eof()) {
3321     if (FormatTok->is(tok::semi))
3322       return;
3323     if (Line->Tokens.empty()) {
3324       // Common issue: Automatic Semicolon Insertion wrapped the line, so the
3325       // import statement should terminate.
3326       return;
3327     }
3328     if (FormatTok->is(tok::l_brace)) {
3329       FormatTok->setBlockKind(BK_Block);
3330       nextToken();
3331       parseBracedList();
3332     } else {
3333       nextToken();
3334     }
3335   }
3336 }
3337 
3338 void UnwrappedLineParser::parseStatementMacro() {
3339   nextToken();
3340   if (FormatTok->is(tok::l_paren))
3341     parseParens();
3342   if (FormatTok->is(tok::semi))
3343     nextToken();
3344   addUnwrappedLine();
3345 }
3346 
3347 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line,
3348                                                  StringRef Prefix = "") {
3349   llvm::dbgs() << Prefix << "Line(" << Line.Level
3350                << ", FSC=" << Line.FirstStartColumn << ")"
3351                << (Line.InPPDirective ? " MACRO" : "") << ": ";
3352   for (const auto &Node : Line.Tokens) {
3353     llvm::dbgs() << Node.Tok->Tok.getName() << "["
3354                  << "T=" << static_cast<unsigned>(Node.Tok->getType())
3355                  << ", OC=" << Node.Tok->OriginalColumn << "] ";
3356   }
3357   for (const auto &Node : Line.Tokens)
3358     for (const auto &ChildNode : Node.Children)
3359       printDebugInfo(ChildNode, "\nChild: ");
3360 
3361   llvm::dbgs() << "\n";
3362 }
3363 
3364 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) {
3365   if (Line->Tokens.empty())
3366     return;
3367   LLVM_DEBUG({
3368     if (CurrentLines == &Lines)
3369       printDebugInfo(*Line);
3370   });
3371 
3372   // If this line closes a block when in Whitesmiths mode, remember that
3373   // information so that the level can be decreased after the line is added.
3374   // This has to happen after the addition of the line since the line itself
3375   // needs to be indented.
3376   bool ClosesWhitesmithsBlock =
3377       Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex &&
3378       Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
3379 
3380   CurrentLines->push_back(std::move(*Line));
3381   Line->Tokens.clear();
3382   Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex;
3383   Line->FirstStartColumn = 0;
3384 
3385   if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove)
3386     --Line->Level;
3387   if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) {
3388     CurrentLines->append(
3389         std::make_move_iterator(PreprocessorDirectives.begin()),
3390         std::make_move_iterator(PreprocessorDirectives.end()));
3391     PreprocessorDirectives.clear();
3392   }
3393   // Disconnect the current token from the last token on the previous line.
3394   FormatTok->Previous = nullptr;
3395 }
3396 
3397 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); }
3398 
3399 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) {
3400   return (Line->InPPDirective || FormatTok.HasUnescapedNewline) &&
3401          FormatTok.NewlinesBefore > 0;
3402 }
3403 
3404 // Checks if \p FormatTok is a line comment that continues the line comment
3405 // section on \p Line.
3406 static bool
3407 continuesLineCommentSection(const FormatToken &FormatTok,
3408                             const UnwrappedLine &Line,
3409                             const llvm::Regex &CommentPragmasRegex) {
3410   if (Line.Tokens.empty())
3411     return false;
3412 
3413   StringRef IndentContent = FormatTok.TokenText;
3414   if (FormatTok.TokenText.startswith("//") ||
3415       FormatTok.TokenText.startswith("/*"))
3416     IndentContent = FormatTok.TokenText.substr(2);
3417   if (CommentPragmasRegex.match(IndentContent))
3418     return false;
3419 
3420   // If Line starts with a line comment, then FormatTok continues the comment
3421   // section if its original column is greater or equal to the original start
3422   // column of the line.
3423   //
3424   // Define the min column token of a line as follows: if a line ends in '{' or
3425   // contains a '{' followed by a line comment, then the min column token is
3426   // that '{'. Otherwise, the min column token of the line is the first token of
3427   // the line.
3428   //
3429   // If Line starts with a token other than a line comment, then FormatTok
3430   // continues the comment section if its original column is greater than the
3431   // original start column of the min column token of the line.
3432   //
3433   // For example, the second line comment continues the first in these cases:
3434   //
3435   // // first line
3436   // // second line
3437   //
3438   // and:
3439   //
3440   // // first line
3441   //  // second line
3442   //
3443   // and:
3444   //
3445   // int i; // first line
3446   //  // second line
3447   //
3448   // and:
3449   //
3450   // do { // first line
3451   //      // second line
3452   //   int i;
3453   // } while (true);
3454   //
3455   // and:
3456   //
3457   // enum {
3458   //   a, // first line
3459   //    // second line
3460   //   b
3461   // };
3462   //
3463   // The second line comment doesn't continue the first in these cases:
3464   //
3465   //   // first line
3466   //  // second line
3467   //
3468   // and:
3469   //
3470   // int i; // first line
3471   // // second line
3472   //
3473   // and:
3474   //
3475   // do { // first line
3476   //   // second line
3477   //   int i;
3478   // } while (true);
3479   //
3480   // and:
3481   //
3482   // enum {
3483   //   a, // first line
3484   //   // second line
3485   // };
3486   const FormatToken *MinColumnToken = Line.Tokens.front().Tok;
3487 
3488   // Scan for '{//'. If found, use the column of '{' as a min column for line
3489   // comment section continuation.
3490   const FormatToken *PreviousToken = nullptr;
3491   for (const UnwrappedLineNode &Node : Line.Tokens) {
3492     if (PreviousToken && PreviousToken->is(tok::l_brace) &&
3493         isLineComment(*Node.Tok)) {
3494       MinColumnToken = PreviousToken;
3495       break;
3496     }
3497     PreviousToken = Node.Tok;
3498 
3499     // Grab the last newline preceding a token in this unwrapped line.
3500     if (Node.Tok->NewlinesBefore > 0) {
3501       MinColumnToken = Node.Tok;
3502     }
3503   }
3504   if (PreviousToken && PreviousToken->is(tok::l_brace)) {
3505     MinColumnToken = PreviousToken;
3506   }
3507 
3508   return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok,
3509                               MinColumnToken);
3510 }
3511 
3512 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
3513   bool JustComments = Line->Tokens.empty();
3514   for (FormatToken *Tok : CommentsBeforeNextToken) {
3515     // Line comments that belong to the same line comment section are put on the
3516     // same line since later we might want to reflow content between them.
3517     // Additional fine-grained breaking of line comment sections is controlled
3518     // by the class BreakableLineCommentSection in case it is desirable to keep
3519     // several line comment sections in the same unwrapped line.
3520     //
3521     // FIXME: Consider putting separate line comment sections as children to the
3522     // unwrapped line instead.
3523     Tok->ContinuesLineCommentSection =
3524         continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex);
3525     if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection)
3526       addUnwrappedLine();
3527     pushToken(Tok);
3528   }
3529   if (NewlineBeforeNext && JustComments)
3530     addUnwrappedLine();
3531   CommentsBeforeNextToken.clear();
3532 }
3533 
3534 void UnwrappedLineParser::nextToken(int LevelDifference) {
3535   if (eof())
3536     return;
3537   flushComments(isOnNewLine(*FormatTok));
3538   pushToken(FormatTok);
3539   FormatToken *Previous = FormatTok;
3540   if (!Style.isJavaScript())
3541     readToken(LevelDifference);
3542   else
3543     readTokenWithJavaScriptASI();
3544   FormatTok->Previous = Previous;
3545 }
3546 
3547 void UnwrappedLineParser::distributeComments(
3548     const SmallVectorImpl<FormatToken *> &Comments,
3549     const FormatToken *NextTok) {
3550   // Whether or not a line comment token continues a line is controlled by
3551   // the method continuesLineCommentSection, with the following caveat:
3552   //
3553   // Define a trail of Comments to be a nonempty proper postfix of Comments such
3554   // that each comment line from the trail is aligned with the next token, if
3555   // the next token exists. If a trail exists, the beginning of the maximal
3556   // trail is marked as a start of a new comment section.
3557   //
3558   // For example in this code:
3559   //
3560   // int a; // line about a
3561   //   // line 1 about b
3562   //   // line 2 about b
3563   //   int b;
3564   //
3565   // the two lines about b form a maximal trail, so there are two sections, the
3566   // first one consisting of the single comment "// line about a" and the
3567   // second one consisting of the next two comments.
3568   if (Comments.empty())
3569     return;
3570   bool ShouldPushCommentsInCurrentLine = true;
3571   bool HasTrailAlignedWithNextToken = false;
3572   unsigned StartOfTrailAlignedWithNextToken = 0;
3573   if (NextTok) {
3574     // We are skipping the first element intentionally.
3575     for (unsigned i = Comments.size() - 1; i > 0; --i) {
3576       if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) {
3577         HasTrailAlignedWithNextToken = true;
3578         StartOfTrailAlignedWithNextToken = i;
3579       }
3580     }
3581   }
3582   for (unsigned i = 0, e = Comments.size(); i < e; ++i) {
3583     FormatToken *FormatTok = Comments[i];
3584     if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) {
3585       FormatTok->ContinuesLineCommentSection = false;
3586     } else {
3587       FormatTok->ContinuesLineCommentSection =
3588           continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex);
3589     }
3590     if (!FormatTok->ContinuesLineCommentSection &&
3591         (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) {
3592       ShouldPushCommentsInCurrentLine = false;
3593     }
3594     if (ShouldPushCommentsInCurrentLine) {
3595       pushToken(FormatTok);
3596     } else {
3597       CommentsBeforeNextToken.push_back(FormatTok);
3598     }
3599   }
3600 }
3601 
3602 void UnwrappedLineParser::readToken(int LevelDifference) {
3603   SmallVector<FormatToken *, 1> Comments;
3604   do {
3605     FormatTok = Tokens->getNextToken();
3606     assert(FormatTok);
3607     while (FormatTok->getType() == TT_ConflictStart ||
3608            FormatTok->getType() == TT_ConflictEnd ||
3609            FormatTok->getType() == TT_ConflictAlternative) {
3610       if (FormatTok->getType() == TT_ConflictStart) {
3611         conditionalCompilationStart(/*Unreachable=*/false);
3612       } else if (FormatTok->getType() == TT_ConflictAlternative) {
3613         conditionalCompilationAlternative();
3614       } else if (FormatTok->getType() == TT_ConflictEnd) {
3615         conditionalCompilationEnd();
3616       }
3617       FormatTok = Tokens->getNextToken();
3618       FormatTok->MustBreakBefore = true;
3619     }
3620 
3621     while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) &&
3622            (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) {
3623       distributeComments(Comments, FormatTok);
3624       Comments.clear();
3625       // If there is an unfinished unwrapped line, we flush the preprocessor
3626       // directives only after that unwrapped line was finished later.
3627       bool SwitchToPreprocessorLines = !Line->Tokens.empty();
3628       ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
3629       assert((LevelDifference >= 0 ||
3630               static_cast<unsigned>(-LevelDifference) <= Line->Level) &&
3631              "LevelDifference makes Line->Level negative");
3632       Line->Level += LevelDifference;
3633       // Comments stored before the preprocessor directive need to be output
3634       // before the preprocessor directive, at the same level as the
3635       // preprocessor directive, as we consider them to apply to the directive.
3636       if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash &&
3637           PPBranchLevel > 0)
3638         Line->Level += PPBranchLevel;
3639       flushComments(isOnNewLine(*FormatTok));
3640       parsePPDirective();
3641     }
3642 
3643     if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) &&
3644         !Line->InPPDirective) {
3645       continue;
3646     }
3647 
3648     if (!FormatTok->Tok.is(tok::comment)) {
3649       distributeComments(Comments, FormatTok);
3650       Comments.clear();
3651       return;
3652     }
3653 
3654     Comments.push_back(FormatTok);
3655   } while (!eof());
3656 
3657   distributeComments(Comments, nullptr);
3658   Comments.clear();
3659 }
3660 
3661 void UnwrappedLineParser::pushToken(FormatToken *Tok) {
3662   Line->Tokens.push_back(UnwrappedLineNode(Tok));
3663   if (MustBreakBeforeNextToken) {
3664     Line->Tokens.back().Tok->MustBreakBefore = true;
3665     MustBreakBeforeNextToken = false;
3666   }
3667 }
3668 
3669 } // end namespace format
3670 } // end namespace clang
3671