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