xref: /netbsd-src/external/apache2/llvm/dist/clang/lib/Format/UnwrappedLineParser.cpp (revision e038c9c4676b0f19b1b7dd08a940c6ed64a6d5ae)
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:
~FormatTokenSource()30   virtual ~FormatTokenSource() {}
31   virtual FormatToken *getNextToken() = 0;
32 
33   virtual unsigned getPosition() = 0;
34   virtual FormatToken *setPosition(unsigned Position) = 0;
35 };
36 
37 namespace {
38 
39 class ScopedDeclarationState {
40 public:
ScopedDeclarationState(UnwrappedLine & Line,std::vector<bool> & Stack,bool MustBeDeclaration)41   ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack,
42                          bool MustBeDeclaration)
43       : Line(Line), Stack(Stack) {
44     Line.MustBeDeclaration = MustBeDeclaration;
45     Stack.push_back(MustBeDeclaration);
46   }
~ScopedDeclarationState()47   ~ScopedDeclarationState() {
48     Stack.pop_back();
49     if (!Stack.empty())
50       Line.MustBeDeclaration = Stack.back();
51     else
52       Line.MustBeDeclaration = true;
53   }
54 
55 private:
56   UnwrappedLine &Line;
57   std::vector<bool> &Stack;
58 };
59 
isLineComment(const FormatToken & FormatTok)60 static bool isLineComment(const FormatToken &FormatTok) {
61   return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*");
62 }
63 
64 // Checks if \p FormatTok is a line comment that continues the line comment
65 // \p Previous. The original column of \p MinColumnToken is used to determine
66 // whether \p FormatTok is indented enough to the right to continue \p Previous.
continuesLineComment(const FormatToken & FormatTok,const FormatToken * Previous,const FormatToken * MinColumnToken)67 static bool continuesLineComment(const FormatToken &FormatTok,
68                                  const FormatToken *Previous,
69                                  const FormatToken *MinColumnToken) {
70   if (!Previous || !MinColumnToken)
71     return false;
72   unsigned MinContinueColumn =
73       MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1);
74   return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 &&
75          isLineComment(*Previous) &&
76          FormatTok.OriginalColumn >= MinContinueColumn;
77 }
78 
79 class ScopedMacroState : public FormatTokenSource {
80 public:
ScopedMacroState(UnwrappedLine & Line,FormatTokenSource * & TokenSource,FormatToken * & ResetToken)81   ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource,
82                    FormatToken *&ResetToken)
83       : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken),
84         PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource),
85         Token(nullptr), PreviousToken(nullptr) {
86     FakeEOF.Tok.startToken();
87     FakeEOF.Tok.setKind(tok::eof);
88     TokenSource = this;
89     Line.Level = 0;
90     Line.InPPDirective = true;
91   }
92 
~ScopedMacroState()93   ~ScopedMacroState() override {
94     TokenSource = PreviousTokenSource;
95     ResetToken = Token;
96     Line.InPPDirective = false;
97     Line.Level = PreviousLineLevel;
98   }
99 
getNextToken()100   FormatToken *getNextToken() override {
101     // The \c UnwrappedLineParser guards against this by never calling
102     // \c getNextToken() after it has encountered the first eof token.
103     assert(!eof());
104     PreviousToken = Token;
105     Token = PreviousTokenSource->getNextToken();
106     if (eof())
107       return &FakeEOF;
108     return Token;
109   }
110 
getPosition()111   unsigned getPosition() override { return PreviousTokenSource->getPosition(); }
112 
setPosition(unsigned Position)113   FormatToken *setPosition(unsigned Position) override {
114     PreviousToken = nullptr;
115     Token = PreviousTokenSource->setPosition(Position);
116     return Token;
117   }
118 
119 private:
eof()120   bool eof() {
121     return Token && Token->HasUnescapedNewline &&
122            !continuesLineComment(*Token, PreviousToken,
123                                  /*MinColumnToken=*/PreviousToken);
124   }
125 
126   FormatToken FakeEOF;
127   UnwrappedLine &Line;
128   FormatTokenSource *&TokenSource;
129   FormatToken *&ResetToken;
130   unsigned PreviousLineLevel;
131   FormatTokenSource *PreviousTokenSource;
132 
133   FormatToken *Token;
134   FormatToken *PreviousToken;
135 };
136 
137 } // end anonymous namespace
138 
139 class ScopedLineState {
140 public:
ScopedLineState(UnwrappedLineParser & Parser,bool SwitchToPreprocessorLines=false)141   ScopedLineState(UnwrappedLineParser &Parser,
142                   bool SwitchToPreprocessorLines = false)
143       : Parser(Parser), OriginalLines(Parser.CurrentLines) {
144     if (SwitchToPreprocessorLines)
145       Parser.CurrentLines = &Parser.PreprocessorDirectives;
146     else if (!Parser.Line->Tokens.empty())
147       Parser.CurrentLines = &Parser.Line->Tokens.back().Children;
148     PreBlockLine = std::move(Parser.Line);
149     Parser.Line = std::make_unique<UnwrappedLine>();
150     Parser.Line->Level = PreBlockLine->Level;
151     Parser.Line->InPPDirective = PreBlockLine->InPPDirective;
152   }
153 
~ScopedLineState()154   ~ScopedLineState() {
155     if (!Parser.Line->Tokens.empty()) {
156       Parser.addUnwrappedLine();
157     }
158     assert(Parser.Line->Tokens.empty());
159     Parser.Line = std::move(PreBlockLine);
160     if (Parser.CurrentLines == &Parser.PreprocessorDirectives)
161       Parser.MustBreakBeforeNextToken = true;
162     Parser.CurrentLines = OriginalLines;
163   }
164 
165 private:
166   UnwrappedLineParser &Parser;
167 
168   std::unique_ptr<UnwrappedLine> PreBlockLine;
169   SmallVectorImpl<UnwrappedLine> *OriginalLines;
170 };
171 
172 class CompoundStatementIndenter {
173 public:
CompoundStatementIndenter(UnwrappedLineParser * Parser,const FormatStyle & Style,unsigned & LineLevel)174   CompoundStatementIndenter(UnwrappedLineParser *Parser,
175                             const FormatStyle &Style, unsigned &LineLevel)
176       : CompoundStatementIndenter(Parser, LineLevel,
177                                   Style.BraceWrapping.AfterControlStatement,
178                                   Style.BraceWrapping.IndentBraces) {}
CompoundStatementIndenter(UnwrappedLineParser * Parser,unsigned & LineLevel,bool WrapBrace,bool IndentBrace)179   CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel,
180                             bool WrapBrace, bool IndentBrace)
181       : LineLevel(LineLevel), OldLineLevel(LineLevel) {
182     if (WrapBrace)
183       Parser->addUnwrappedLine();
184     if (IndentBrace)
185       ++LineLevel;
186   }
~CompoundStatementIndenter()187   ~CompoundStatementIndenter() { LineLevel = OldLineLevel; }
188 
189 private:
190   unsigned &LineLevel;
191   unsigned OldLineLevel;
192 };
193 
194 namespace {
195 
196 class IndexedTokenSource : public FormatTokenSource {
197 public:
IndexedTokenSource(ArrayRef<FormatToken * > Tokens)198   IndexedTokenSource(ArrayRef<FormatToken *> Tokens)
199       : Tokens(Tokens), Position(-1) {}
200 
getNextToken()201   FormatToken *getNextToken() override {
202     ++Position;
203     return Tokens[Position];
204   }
205 
getPosition()206   unsigned getPosition() override {
207     assert(Position >= 0);
208     return Position;
209   }
210 
setPosition(unsigned P)211   FormatToken *setPosition(unsigned P) override {
212     Position = P;
213     return Tokens[Position];
214   }
215 
reset()216   void reset() { Position = -1; }
217 
218 private:
219   ArrayRef<FormatToken *> Tokens;
220   int Position;
221 };
222 
223 } // end anonymous namespace
224 
UnwrappedLineParser(const FormatStyle & Style,const AdditionalKeywords & Keywords,unsigned FirstStartColumn,ArrayRef<FormatToken * > Tokens,UnwrappedLineConsumer & Callback)225 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style,
226                                          const AdditionalKeywords &Keywords,
227                                          unsigned FirstStartColumn,
228                                          ArrayRef<FormatToken *> Tokens,
229                                          UnwrappedLineConsumer &Callback)
230     : Line(new UnwrappedLine), MustBreakBeforeNextToken(false),
231       CurrentLines(&Lines), Style(Style), Keywords(Keywords),
232       CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr),
233       Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1),
234       IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None
235                        ? IG_Rejected
236                        : IG_Inited),
237       IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn) {}
238 
reset()239 void UnwrappedLineParser::reset() {
240   PPBranchLevel = -1;
241   IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None
242                      ? IG_Rejected
243                      : IG_Inited;
244   IncludeGuardToken = nullptr;
245   Line.reset(new UnwrappedLine);
246   CommentsBeforeNextToken.clear();
247   FormatTok = nullptr;
248   MustBreakBeforeNextToken = false;
249   PreprocessorDirectives.clear();
250   CurrentLines = &Lines;
251   DeclarationScopeStack.clear();
252   PPStack.clear();
253   Line->FirstStartColumn = FirstStartColumn;
254 }
255 
parse()256 void UnwrappedLineParser::parse() {
257   IndexedTokenSource TokenSource(AllTokens);
258   Line->FirstStartColumn = FirstStartColumn;
259   do {
260     LLVM_DEBUG(llvm::dbgs() << "----\n");
261     reset();
262     Tokens = &TokenSource;
263     TokenSource.reset();
264 
265     readToken();
266     parseFile();
267 
268     // If we found an include guard then all preprocessor directives (other than
269     // the guard) are over-indented by one.
270     if (IncludeGuard == IG_Found)
271       for (auto &Line : Lines)
272         if (Line.InPPDirective && Line.Level > 0)
273           --Line.Level;
274 
275     // Create line with eof token.
276     pushToken(FormatTok);
277     addUnwrappedLine();
278 
279     for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(),
280                                                   E = Lines.end();
281          I != E; ++I) {
282       Callback.consumeUnwrappedLine(*I);
283     }
284     Callback.finishRun();
285     Lines.clear();
286     while (!PPLevelBranchIndex.empty() &&
287            PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) {
288       PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1);
289       PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1);
290     }
291     if (!PPLevelBranchIndex.empty()) {
292       ++PPLevelBranchIndex.back();
293       assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size());
294       assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back());
295     }
296   } while (!PPLevelBranchIndex.empty());
297 }
298 
parseFile()299 void UnwrappedLineParser::parseFile() {
300   // The top-level context in a file always has declarations, except for pre-
301   // processor directives and JavaScript files.
302   bool MustBeDeclaration =
303       !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript;
304   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
305                                           MustBeDeclaration);
306   if (Style.Language == FormatStyle::LK_TextProto)
307     parseBracedList();
308   else
309     parseLevel(/*HasOpeningBrace=*/false);
310   // Make sure to format the remaining tokens.
311   //
312   // LK_TextProto is special since its top-level is parsed as the body of a
313   // braced list, which does not necessarily have natural line separators such
314   // as a semicolon. Comments after the last entry that have been determined to
315   // not belong to that line, as in:
316   //   key: value
317   //   // endfile comment
318   // do not have a chance to be put on a line of their own until this point.
319   // Here we add this newline before end-of-file comments.
320   if (Style.Language == FormatStyle::LK_TextProto &&
321       !CommentsBeforeNextToken.empty())
322     addUnwrappedLine();
323   flushComments(true);
324   addUnwrappedLine();
325 }
326 
parseCSharpGenericTypeConstraint()327 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() {
328   do {
329     switch (FormatTok->Tok.getKind()) {
330     case tok::l_brace:
331       return;
332     default:
333       if (FormatTok->is(Keywords.kw_where)) {
334         addUnwrappedLine();
335         nextToken();
336         parseCSharpGenericTypeConstraint();
337         break;
338       }
339       nextToken();
340       break;
341     }
342   } while (!eof());
343 }
344 
parseCSharpAttribute()345 void UnwrappedLineParser::parseCSharpAttribute() {
346   int UnpairedSquareBrackets = 1;
347   do {
348     switch (FormatTok->Tok.getKind()) {
349     case tok::r_square:
350       nextToken();
351       --UnpairedSquareBrackets;
352       if (UnpairedSquareBrackets == 0) {
353         addUnwrappedLine();
354         return;
355       }
356       break;
357     case tok::l_square:
358       ++UnpairedSquareBrackets;
359       nextToken();
360       break;
361     default:
362       nextToken();
363       break;
364     }
365   } while (!eof());
366 }
367 
parseLevel(bool HasOpeningBrace)368 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) {
369   bool SwitchLabelEncountered = false;
370   do {
371     tok::TokenKind kind = FormatTok->Tok.getKind();
372     if (FormatTok->getType() == TT_MacroBlockBegin) {
373       kind = tok::l_brace;
374     } else if (FormatTok->getType() == TT_MacroBlockEnd) {
375       kind = tok::r_brace;
376     }
377 
378     switch (kind) {
379     case tok::comment:
380       nextToken();
381       addUnwrappedLine();
382       break;
383     case tok::l_brace:
384       // FIXME: Add parameter whether this can happen - if this happens, we must
385       // be in a non-declaration context.
386       if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList())
387         continue;
388       parseBlock(/*MustBeDeclaration=*/false);
389       addUnwrappedLine();
390       break;
391     case tok::r_brace:
392       if (HasOpeningBrace)
393         return;
394       nextToken();
395       addUnwrappedLine();
396       break;
397     case tok::kw_default: {
398       unsigned StoredPosition = Tokens->getPosition();
399       FormatToken *Next;
400       do {
401         Next = Tokens->getNextToken();
402       } while (Next && Next->is(tok::comment));
403       FormatTok = Tokens->setPosition(StoredPosition);
404       if (Next && Next->isNot(tok::colon)) {
405         // default not followed by ':' is not a case label; treat it like
406         // an identifier.
407         parseStructuralElement();
408         break;
409       }
410       // Else, if it is 'default:', fall through to the case handling.
411       LLVM_FALLTHROUGH;
412     }
413     case tok::kw_case:
414       if (Style.Language == FormatStyle::LK_JavaScript &&
415           Line->MustBeDeclaration) {
416         // A 'case: string' style field declaration.
417         parseStructuralElement();
418         break;
419       }
420       if (!SwitchLabelEncountered &&
421           (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1)))
422         ++Line->Level;
423       SwitchLabelEncountered = true;
424       parseStructuralElement();
425       break;
426     case tok::l_square:
427       if (Style.isCSharp()) {
428         nextToken();
429         parseCSharpAttribute();
430         break;
431       }
432       LLVM_FALLTHROUGH;
433     default:
434       parseStructuralElement();
435       break;
436     }
437   } while (!eof());
438 }
439 
calculateBraceTypes(bool ExpectClassBody)440 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
441   // We'll parse forward through the tokens until we hit
442   // a closing brace or eof - note that getNextToken() will
443   // parse macros, so this will magically work inside macro
444   // definitions, too.
445   unsigned StoredPosition = Tokens->getPosition();
446   FormatToken *Tok = FormatTok;
447   const FormatToken *PrevTok = Tok->Previous;
448   // Keep a stack of positions of lbrace tokens. We will
449   // update information about whether an lbrace starts a
450   // braced init list or a different block during the loop.
451   SmallVector<FormatToken *, 8> LBraceStack;
452   assert(Tok->Tok.is(tok::l_brace));
453   do {
454     // Get next non-comment token.
455     FormatToken *NextTok;
456     unsigned ReadTokens = 0;
457     do {
458       NextTok = Tokens->getNextToken();
459       ++ReadTokens;
460     } while (NextTok->is(tok::comment));
461 
462     switch (Tok->Tok.getKind()) {
463     case tok::l_brace:
464       if (Style.Language == FormatStyle::LK_JavaScript && PrevTok) {
465         if (PrevTok->isOneOf(tok::colon, tok::less))
466           // A ':' indicates this code is in a type, or a braced list
467           // following a label in an object literal ({a: {b: 1}}).
468           // A '<' could be an object used in a comparison, but that is nonsense
469           // code (can never return true), so more likely it is a generic type
470           // argument (`X<{a: string; b: number}>`).
471           // The code below could be confused by semicolons between the
472           // individual members in a type member list, which would normally
473           // trigger BK_Block. In both cases, this must be parsed as an inline
474           // braced init.
475           Tok->setBlockKind(BK_BracedInit);
476         else if (PrevTok->is(tok::r_paren))
477           // `) { }` can only occur in function or method declarations in JS.
478           Tok->setBlockKind(BK_Block);
479       } else {
480         Tok->setBlockKind(BK_Unknown);
481       }
482       LBraceStack.push_back(Tok);
483       break;
484     case tok::r_brace:
485       if (LBraceStack.empty())
486         break;
487       if (LBraceStack.back()->is(BK_Unknown)) {
488         bool ProbablyBracedList = false;
489         if (Style.Language == FormatStyle::LK_Proto) {
490           ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square);
491         } else {
492           // Using OriginalColumn to distinguish between ObjC methods and
493           // binary operators is a bit hacky.
494           bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) &&
495                                   NextTok->OriginalColumn == 0;
496 
497           // If there is a comma, semicolon or right paren after the closing
498           // brace, we assume this is a braced initializer list.  Note that
499           // regardless how we mark inner braces here, we will overwrite the
500           // BlockKind later if we parse a braced list (where all blocks
501           // inside are by default braced lists), or when we explicitly detect
502           // blocks (for example while parsing lambdas).
503           // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a
504           // braced list in JS.
505           ProbablyBracedList =
506               (Style.Language == FormatStyle::LK_JavaScript &&
507                NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in,
508                                 Keywords.kw_as)) ||
509               (Style.isCpp() && NextTok->is(tok::l_paren)) ||
510               NextTok->isOneOf(tok::comma, tok::period, tok::colon,
511                                tok::r_paren, tok::r_square, tok::l_brace,
512                                tok::ellipsis) ||
513               (NextTok->is(tok::identifier) &&
514                !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) ||
515               (NextTok->is(tok::semi) &&
516                (!ExpectClassBody || LBraceStack.size() != 1)) ||
517               (NextTok->isBinaryOperator() && !NextIsObjCMethod);
518           if (!Style.isCSharp() && NextTok->is(tok::l_square)) {
519             // We can have an array subscript after a braced init
520             // list, but C++11 attributes are expected after blocks.
521             NextTok = Tokens->getNextToken();
522             ++ReadTokens;
523             ProbablyBracedList = NextTok->isNot(tok::l_square);
524           }
525         }
526         if (ProbablyBracedList) {
527           Tok->setBlockKind(BK_BracedInit);
528           LBraceStack.back()->setBlockKind(BK_BracedInit);
529         } else {
530           Tok->setBlockKind(BK_Block);
531           LBraceStack.back()->setBlockKind(BK_Block);
532         }
533       }
534       LBraceStack.pop_back();
535       break;
536     case tok::identifier:
537       if (!Tok->is(TT_StatementMacro))
538         break;
539       LLVM_FALLTHROUGH;
540     case tok::at:
541     case tok::semi:
542     case tok::kw_if:
543     case tok::kw_while:
544     case tok::kw_for:
545     case tok::kw_switch:
546     case tok::kw_try:
547     case tok::kw___try:
548       if (!LBraceStack.empty() && LBraceStack.back()->is(BK_Unknown))
549         LBraceStack.back()->setBlockKind(BK_Block);
550       break;
551     default:
552       break;
553     }
554     PrevTok = Tok;
555     Tok = NextTok;
556   } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty());
557 
558   // Assume other blocks for all unclosed opening braces.
559   for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) {
560     if (LBraceStack[i]->is(BK_Unknown))
561       LBraceStack[i]->setBlockKind(BK_Block);
562   }
563 
564   FormatTok = Tokens->setPosition(StoredPosition);
565 }
566 
567 template <class T>
hash_combine(std::size_t & seed,const T & v)568 static inline void hash_combine(std::size_t &seed, const T &v) {
569   std::hash<T> hasher;
570   seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
571 }
572 
computePPHash() const573 size_t UnwrappedLineParser::computePPHash() const {
574   size_t h = 0;
575   for (const auto &i : PPStack) {
576     hash_combine(h, size_t(i.Kind));
577     hash_combine(h, i.Line);
578   }
579   return h;
580 }
581 
parseBlock(bool MustBeDeclaration,unsigned AddLevels,bool MunchSemi,bool UnindentWhitesmithsBraces)582 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, unsigned AddLevels,
583                                      bool MunchSemi,
584                                      bool UnindentWhitesmithsBraces) {
585   assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) &&
586          "'{' or macro block token expected");
587   const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin);
588   FormatTok->setBlockKind(BK_Block);
589 
590   // For Whitesmiths mode, jump to the next level prior to skipping over the
591   // braces.
592   if (AddLevels > 0 && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
593     ++Line->Level;
594 
595   size_t PPStartHash = computePPHash();
596 
597   unsigned InitialLevel = Line->Level;
598   nextToken(/*LevelDifference=*/AddLevels);
599 
600   if (MacroBlock && FormatTok->is(tok::l_paren))
601     parseParens();
602 
603   size_t NbPreprocessorDirectives =
604       CurrentLines == &Lines ? PreprocessorDirectives.size() : 0;
605   addUnwrappedLine();
606   size_t OpeningLineIndex =
607       CurrentLines->empty()
608           ? (UnwrappedLine::kInvalidIndex)
609           : (CurrentLines->size() - 1 - NbPreprocessorDirectives);
610 
611   // Whitesmiths is weird here. The brace needs to be indented for the namespace
612   // block, but the block itself may not be indented depending on the style
613   // settings. This allows the format to back up one level in those cases.
614   if (UnindentWhitesmithsBraces)
615     --Line->Level;
616 
617   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
618                                           MustBeDeclaration);
619   if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths)
620     Line->Level += AddLevels;
621   parseLevel(/*HasOpeningBrace=*/true);
622 
623   if (eof())
624     return;
625 
626   if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd)
627                  : !FormatTok->is(tok::r_brace)) {
628     Line->Level = InitialLevel;
629     FormatTok->setBlockKind(BK_Block);
630     return;
631   }
632 
633   size_t PPEndHash = computePPHash();
634 
635   // Munch the closing brace.
636   nextToken(/*LevelDifference=*/-AddLevels);
637 
638   if (MacroBlock && FormatTok->is(tok::l_paren))
639     parseParens();
640 
641   if (FormatTok->is(tok::arrow)) {
642     // Following the } we can find a trailing return type arrow
643     // as part of an implicit conversion constraint.
644     nextToken();
645     parseStructuralElement();
646   }
647 
648   if (MunchSemi && FormatTok->Tok.is(tok::semi))
649     nextToken();
650 
651   Line->Level = InitialLevel;
652 
653   if (PPStartHash == PPEndHash) {
654     Line->MatchingOpeningBlockLineIndex = OpeningLineIndex;
655     if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) {
656       // Update the opening line to add the forward reference as well
657       (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex =
658           CurrentLines->size() - 1;
659     }
660   }
661 }
662 
isGoogScope(const UnwrappedLine & Line)663 static bool isGoogScope(const UnwrappedLine &Line) {
664   // FIXME: Closure-library specific stuff should not be hard-coded but be
665   // configurable.
666   if (Line.Tokens.size() < 4)
667     return false;
668   auto I = Line.Tokens.begin();
669   if (I->Tok->TokenText != "goog")
670     return false;
671   ++I;
672   if (I->Tok->isNot(tok::period))
673     return false;
674   ++I;
675   if (I->Tok->TokenText != "scope")
676     return false;
677   ++I;
678   return I->Tok->is(tok::l_paren);
679 }
680 
isIIFE(const UnwrappedLine & Line,const AdditionalKeywords & Keywords)681 static bool isIIFE(const UnwrappedLine &Line,
682                    const AdditionalKeywords &Keywords) {
683   // Look for the start of an immediately invoked anonymous function.
684   // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression
685   // This is commonly done in JavaScript to create a new, anonymous scope.
686   // Example: (function() { ... })()
687   if (Line.Tokens.size() < 3)
688     return false;
689   auto I = Line.Tokens.begin();
690   if (I->Tok->isNot(tok::l_paren))
691     return false;
692   ++I;
693   if (I->Tok->isNot(Keywords.kw_function))
694     return false;
695   ++I;
696   return I->Tok->is(tok::l_paren);
697 }
698 
ShouldBreakBeforeBrace(const FormatStyle & Style,const FormatToken & InitialToken)699 static bool ShouldBreakBeforeBrace(const FormatStyle &Style,
700                                    const FormatToken &InitialToken) {
701   if (InitialToken.isOneOf(tok::kw_namespace, TT_NamespaceMacro))
702     return Style.BraceWrapping.AfterNamespace;
703   if (InitialToken.is(tok::kw_class))
704     return Style.BraceWrapping.AfterClass;
705   if (InitialToken.is(tok::kw_union))
706     return Style.BraceWrapping.AfterUnion;
707   if (InitialToken.is(tok::kw_struct))
708     return Style.BraceWrapping.AfterStruct;
709   return false;
710 }
711 
parseChildBlock()712 void UnwrappedLineParser::parseChildBlock() {
713   FormatTok->setBlockKind(BK_Block);
714   nextToken();
715   {
716     bool SkipIndent = (Style.Language == FormatStyle::LK_JavaScript &&
717                        (isGoogScope(*Line) || isIIFE(*Line, Keywords)));
718     ScopedLineState LineState(*this);
719     ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
720                                             /*MustBeDeclaration=*/false);
721     Line->Level += SkipIndent ? 0 : 1;
722     parseLevel(/*HasOpeningBrace=*/true);
723     flushComments(isOnNewLine(*FormatTok));
724     Line->Level -= SkipIndent ? 0 : 1;
725   }
726   nextToken();
727 }
728 
parsePPDirective()729 void UnwrappedLineParser::parsePPDirective() {
730   assert(FormatTok->Tok.is(tok::hash) && "'#' expected");
731   ScopedMacroState MacroState(*Line, Tokens, FormatTok);
732 
733   nextToken();
734 
735   if (!FormatTok->Tok.getIdentifierInfo()) {
736     parsePPUnknown();
737     return;
738   }
739 
740   switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) {
741   case tok::pp_define:
742     parsePPDefine();
743     return;
744   case tok::pp_if:
745     parsePPIf(/*IfDef=*/false);
746     break;
747   case tok::pp_ifdef:
748   case tok::pp_ifndef:
749     parsePPIf(/*IfDef=*/true);
750     break;
751   case tok::pp_else:
752     parsePPElse();
753     break;
754   case tok::pp_elif:
755     parsePPElIf();
756     break;
757   case tok::pp_endif:
758     parsePPEndIf();
759     break;
760   default:
761     parsePPUnknown();
762     break;
763   }
764 }
765 
conditionalCompilationCondition(bool Unreachable)766 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) {
767   size_t Line = CurrentLines->size();
768   if (CurrentLines == &PreprocessorDirectives)
769     Line += Lines.size();
770 
771   if (Unreachable ||
772       (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable))
773     PPStack.push_back({PP_Unreachable, Line});
774   else
775     PPStack.push_back({PP_Conditional, Line});
776 }
777 
conditionalCompilationStart(bool Unreachable)778 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) {
779   ++PPBranchLevel;
780   assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size());
781   if (PPBranchLevel == (int)PPLevelBranchIndex.size()) {
782     PPLevelBranchIndex.push_back(0);
783     PPLevelBranchCount.push_back(0);
784   }
785   PPChainBranchIndex.push(0);
786   bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0;
787   conditionalCompilationCondition(Unreachable || Skip);
788 }
789 
conditionalCompilationAlternative()790 void UnwrappedLineParser::conditionalCompilationAlternative() {
791   if (!PPStack.empty())
792     PPStack.pop_back();
793   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
794   if (!PPChainBranchIndex.empty())
795     ++PPChainBranchIndex.top();
796   conditionalCompilationCondition(
797       PPBranchLevel >= 0 && !PPChainBranchIndex.empty() &&
798       PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top());
799 }
800 
conditionalCompilationEnd()801 void UnwrappedLineParser::conditionalCompilationEnd() {
802   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
803   if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) {
804     if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) {
805       PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1;
806     }
807   }
808   // Guard against #endif's without #if.
809   if (PPBranchLevel > -1)
810     --PPBranchLevel;
811   if (!PPChainBranchIndex.empty())
812     PPChainBranchIndex.pop();
813   if (!PPStack.empty())
814     PPStack.pop_back();
815 }
816 
parsePPIf(bool IfDef)817 void UnwrappedLineParser::parsePPIf(bool IfDef) {
818   bool IfNDef = FormatTok->is(tok::pp_ifndef);
819   nextToken();
820   bool Unreachable = false;
821   if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0"))
822     Unreachable = true;
823   if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG")
824     Unreachable = true;
825   conditionalCompilationStart(Unreachable);
826   FormatToken *IfCondition = FormatTok;
827   // If there's a #ifndef on the first line, and the only lines before it are
828   // comments, it could be an include guard.
829   bool MaybeIncludeGuard = IfNDef;
830   if (IncludeGuard == IG_Inited && MaybeIncludeGuard)
831     for (auto &Line : Lines) {
832       if (!Line.Tokens.front().Tok->is(tok::comment)) {
833         MaybeIncludeGuard = false;
834         IncludeGuard = IG_Rejected;
835         break;
836       }
837     }
838   --PPBranchLevel;
839   parsePPUnknown();
840   ++PPBranchLevel;
841   if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
842     IncludeGuard = IG_IfNdefed;
843     IncludeGuardToken = IfCondition;
844   }
845 }
846 
parsePPElse()847 void UnwrappedLineParser::parsePPElse() {
848   // If a potential include guard has an #else, it's not an include guard.
849   if (IncludeGuard == IG_Defined && PPBranchLevel == 0)
850     IncludeGuard = IG_Rejected;
851   conditionalCompilationAlternative();
852   if (PPBranchLevel > -1)
853     --PPBranchLevel;
854   parsePPUnknown();
855   ++PPBranchLevel;
856 }
857 
parsePPElIf()858 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); }
859 
parsePPEndIf()860 void UnwrappedLineParser::parsePPEndIf() {
861   conditionalCompilationEnd();
862   parsePPUnknown();
863   // If the #endif of a potential include guard is the last thing in the file,
864   // then we found an include guard.
865   unsigned TokenPosition = Tokens->getPosition();
866   FormatToken *PeekNext = AllTokens[TokenPosition];
867   if (IncludeGuard == IG_Defined && PPBranchLevel == -1 &&
868       PeekNext->is(tok::eof) &&
869       Style.IndentPPDirectives != FormatStyle::PPDIS_None)
870     IncludeGuard = IG_Found;
871 }
872 
parsePPDefine()873 void UnwrappedLineParser::parsePPDefine() {
874   nextToken();
875 
876   if (!FormatTok->Tok.getIdentifierInfo()) {
877     IncludeGuard = IG_Rejected;
878     IncludeGuardToken = nullptr;
879     parsePPUnknown();
880     return;
881   }
882 
883   if (IncludeGuard == IG_IfNdefed &&
884       IncludeGuardToken->TokenText == FormatTok->TokenText) {
885     IncludeGuard = IG_Defined;
886     IncludeGuardToken = nullptr;
887     for (auto &Line : Lines) {
888       if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) {
889         IncludeGuard = IG_Rejected;
890         break;
891       }
892     }
893   }
894 
895   nextToken();
896   if (FormatTok->Tok.getKind() == tok::l_paren &&
897       FormatTok->WhitespaceRange.getBegin() ==
898           FormatTok->WhitespaceRange.getEnd()) {
899     parseParens();
900   }
901   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
902     Line->Level += PPBranchLevel + 1;
903   addUnwrappedLine();
904   ++Line->Level;
905 
906   // Errors during a preprocessor directive can only affect the layout of the
907   // preprocessor directive, and thus we ignore them. An alternative approach
908   // would be to use the same approach we use on the file level (no
909   // re-indentation if there was a structural error) within the macro
910   // definition.
911   parseFile();
912 }
913 
parsePPUnknown()914 void UnwrappedLineParser::parsePPUnknown() {
915   do {
916     nextToken();
917   } while (!eof());
918   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
919     Line->Level += PPBranchLevel + 1;
920   addUnwrappedLine();
921 }
922 
923 // Here we exclude certain tokens that are not usually the first token in an
924 // unwrapped line. This is used in attempt to distinguish macro calls without
925 // trailing semicolons from other constructs split to several lines.
tokenCanStartNewLine(const FormatToken & Tok)926 static bool tokenCanStartNewLine(const FormatToken &Tok) {
927   // Semicolon can be a null-statement, l_square can be a start of a macro or
928   // a C++11 attribute, but this doesn't seem to be common.
929   return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) &&
930          Tok.isNot(TT_AttributeSquare) &&
931          // Tokens that can only be used as binary operators and a part of
932          // overloaded operator names.
933          Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) &&
934          Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) &&
935          Tok.isNot(tok::less) && Tok.isNot(tok::greater) &&
936          Tok.isNot(tok::slash) && Tok.isNot(tok::percent) &&
937          Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) &&
938          Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) &&
939          Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) &&
940          Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) &&
941          Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) &&
942          Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) &&
943          Tok.isNot(tok::lesslessequal) &&
944          // Colon is used in labels, base class lists, initializer lists,
945          // range-based for loops, ternary operator, but should never be the
946          // first token in an unwrapped line.
947          Tok.isNot(tok::colon) &&
948          // 'noexcept' is a trailing annotation.
949          Tok.isNot(tok::kw_noexcept);
950 }
951 
mustBeJSIdent(const AdditionalKeywords & Keywords,const FormatToken * FormatTok)952 static bool mustBeJSIdent(const AdditionalKeywords &Keywords,
953                           const FormatToken *FormatTok) {
954   // FIXME: This returns true for C/C++ keywords like 'struct'.
955   return FormatTok->is(tok::identifier) &&
956          (FormatTok->Tok.getIdentifierInfo() == nullptr ||
957           !FormatTok->isOneOf(
958               Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async,
959               Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally,
960               Keywords.kw_function, Keywords.kw_import, Keywords.kw_is,
961               Keywords.kw_let, Keywords.kw_var, tok::kw_const,
962               Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements,
963               Keywords.kw_instanceof, Keywords.kw_interface, Keywords.kw_throws,
964               Keywords.kw_from));
965 }
966 
mustBeJSIdentOrValue(const AdditionalKeywords & Keywords,const FormatToken * FormatTok)967 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords,
968                                  const FormatToken *FormatTok) {
969   return FormatTok->Tok.isLiteral() ||
970          FormatTok->isOneOf(tok::kw_true, tok::kw_false) ||
971          mustBeJSIdent(Keywords, FormatTok);
972 }
973 
974 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement
975 // when encountered after a value (see mustBeJSIdentOrValue).
isJSDeclOrStmt(const AdditionalKeywords & Keywords,const FormatToken * FormatTok)976 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords,
977                            const FormatToken *FormatTok) {
978   return FormatTok->isOneOf(
979       tok::kw_return, Keywords.kw_yield,
980       // conditionals
981       tok::kw_if, tok::kw_else,
982       // loops
983       tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break,
984       // switch/case
985       tok::kw_switch, tok::kw_case,
986       // exceptions
987       tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally,
988       // declaration
989       tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let,
990       Keywords.kw_async, Keywords.kw_function,
991       // import/export
992       Keywords.kw_import, tok::kw_export);
993 }
994 
995 // readTokenWithJavaScriptASI reads the next token and terminates the current
996 // line if JavaScript Automatic Semicolon Insertion must
997 // happen between the current token and the next token.
998 //
999 // This method is conservative - it cannot cover all edge cases of JavaScript,
1000 // but only aims to correctly handle certain well known cases. It *must not*
1001 // return true in speculative cases.
readTokenWithJavaScriptASI()1002 void UnwrappedLineParser::readTokenWithJavaScriptASI() {
1003   FormatToken *Previous = FormatTok;
1004   readToken();
1005   FormatToken *Next = FormatTok;
1006 
1007   bool IsOnSameLine =
1008       CommentsBeforeNextToken.empty()
1009           ? Next->NewlinesBefore == 0
1010           : CommentsBeforeNextToken.front()->NewlinesBefore == 0;
1011   if (IsOnSameLine)
1012     return;
1013 
1014   bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous);
1015   bool PreviousStartsTemplateExpr =
1016       Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${");
1017   if (PreviousMustBeValue || Previous->is(tok::r_paren)) {
1018     // If the line contains an '@' sign, the previous token might be an
1019     // annotation, which can precede another identifier/value.
1020     bool HasAt = std::find_if(Line->Tokens.begin(), Line->Tokens.end(),
1021                               [](UnwrappedLineNode &LineNode) {
1022                                 return LineNode.Tok->is(tok::at);
1023                               }) != Line->Tokens.end();
1024     if (HasAt)
1025       return;
1026   }
1027   if (Next->is(tok::exclaim) && PreviousMustBeValue)
1028     return addUnwrappedLine();
1029   bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next);
1030   bool NextEndsTemplateExpr =
1031       Next->is(TT_TemplateString) && Next->TokenText.startswith("}");
1032   if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr &&
1033       (PreviousMustBeValue ||
1034        Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus,
1035                          tok::minusminus)))
1036     return addUnwrappedLine();
1037   if ((PreviousMustBeValue || Previous->is(tok::r_paren)) &&
1038       isJSDeclOrStmt(Keywords, Next))
1039     return addUnwrappedLine();
1040 }
1041 
parseStructuralElement()1042 void UnwrappedLineParser::parseStructuralElement() {
1043   assert(!FormatTok->is(tok::l_brace));
1044   if (Style.Language == FormatStyle::LK_TableGen &&
1045       FormatTok->is(tok::pp_include)) {
1046     nextToken();
1047     if (FormatTok->is(tok::string_literal))
1048       nextToken();
1049     addUnwrappedLine();
1050     return;
1051   }
1052   switch (FormatTok->Tok.getKind()) {
1053   case tok::kw_asm:
1054     nextToken();
1055     if (FormatTok->is(tok::l_brace)) {
1056       FormatTok->setType(TT_InlineASMBrace);
1057       nextToken();
1058       while (FormatTok && FormatTok->isNot(tok::eof)) {
1059         if (FormatTok->is(tok::r_brace)) {
1060           FormatTok->setType(TT_InlineASMBrace);
1061           nextToken();
1062           addUnwrappedLine();
1063           break;
1064         }
1065         FormatTok->Finalized = true;
1066         nextToken();
1067       }
1068     }
1069     break;
1070   case tok::kw_namespace:
1071     parseNamespace();
1072     return;
1073   case tok::kw_public:
1074   case tok::kw_protected:
1075   case tok::kw_private:
1076     if (Style.Language == FormatStyle::LK_Java ||
1077         Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp())
1078       nextToken();
1079     else
1080       parseAccessSpecifier();
1081     return;
1082   case tok::kw_if:
1083     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1084       // field/method declaration.
1085       break;
1086     parseIfThenElse();
1087     return;
1088   case tok::kw_for:
1089   case tok::kw_while:
1090     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1091       // field/method declaration.
1092       break;
1093     parseForOrWhileLoop();
1094     return;
1095   case tok::kw_do:
1096     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1097       // field/method declaration.
1098       break;
1099     parseDoWhile();
1100     return;
1101   case tok::kw_switch:
1102     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1103       // 'switch: string' field declaration.
1104       break;
1105     parseSwitch();
1106     return;
1107   case tok::kw_default:
1108     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1109       // 'default: string' field declaration.
1110       break;
1111     nextToken();
1112     if (FormatTok->is(tok::colon)) {
1113       parseLabel();
1114       return;
1115     }
1116     // e.g. "default void f() {}" in a Java interface.
1117     break;
1118   case tok::kw_case:
1119     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1120       // 'case: string' field declaration.
1121       break;
1122     parseCaseLabel();
1123     return;
1124   case tok::kw_try:
1125   case tok::kw___try:
1126     if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration)
1127       // field/method declaration.
1128       break;
1129     parseTryCatch();
1130     return;
1131   case tok::kw_extern:
1132     nextToken();
1133     if (FormatTok->Tok.is(tok::string_literal)) {
1134       nextToken();
1135       if (FormatTok->Tok.is(tok::l_brace)) {
1136         if (!Style.IndentExternBlock) {
1137           if (Style.BraceWrapping.AfterExternBlock) {
1138             addUnwrappedLine();
1139           }
1140           unsigned AddLevels = Style.BraceWrapping.AfterExternBlock ? 1u : 0u;
1141           parseBlock(/*MustBeDeclaration=*/true, AddLevels);
1142         } else {
1143           unsigned AddLevels =
1144               Style.IndentExternBlock == FormatStyle::IEBS_Indent ? 1u : 0u;
1145           parseBlock(/*MustBeDeclaration=*/true, AddLevels);
1146         }
1147         addUnwrappedLine();
1148         return;
1149       }
1150     }
1151     break;
1152   case tok::kw_export:
1153     if (Style.Language == FormatStyle::LK_JavaScript) {
1154       parseJavaScriptEs6ImportExport();
1155       return;
1156     }
1157     if (!Style.isCpp())
1158       break;
1159     // Handle C++ "(inline|export) namespace".
1160     LLVM_FALLTHROUGH;
1161   case tok::kw_inline:
1162     nextToken();
1163     if (FormatTok->Tok.is(tok::kw_namespace)) {
1164       parseNamespace();
1165       return;
1166     }
1167     break;
1168   case tok::identifier:
1169     if (FormatTok->is(TT_ForEachMacro)) {
1170       parseForOrWhileLoop();
1171       return;
1172     }
1173     if (FormatTok->is(TT_MacroBlockBegin)) {
1174       parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
1175                  /*MunchSemi=*/false);
1176       return;
1177     }
1178     if (FormatTok->is(Keywords.kw_import)) {
1179       if (Style.Language == FormatStyle::LK_JavaScript) {
1180         parseJavaScriptEs6ImportExport();
1181         return;
1182       }
1183       if (Style.Language == FormatStyle::LK_Proto) {
1184         nextToken();
1185         if (FormatTok->is(tok::kw_public))
1186           nextToken();
1187         if (!FormatTok->is(tok::string_literal))
1188           return;
1189         nextToken();
1190         if (FormatTok->is(tok::semi))
1191           nextToken();
1192         addUnwrappedLine();
1193         return;
1194       }
1195     }
1196     if (Style.isCpp() &&
1197         FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals,
1198                            Keywords.kw_slots, Keywords.kw_qslots)) {
1199       nextToken();
1200       if (FormatTok->is(tok::colon)) {
1201         nextToken();
1202         addUnwrappedLine();
1203         return;
1204       }
1205     }
1206     if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) {
1207       parseStatementMacro();
1208       return;
1209     }
1210     if (Style.isCpp() && FormatTok->is(TT_NamespaceMacro)) {
1211       parseNamespace();
1212       return;
1213     }
1214     // In all other cases, parse the declaration.
1215     break;
1216   default:
1217     break;
1218   }
1219   do {
1220     const FormatToken *Previous = FormatTok->Previous;
1221     switch (FormatTok->Tok.getKind()) {
1222     case tok::at:
1223       nextToken();
1224       if (FormatTok->Tok.is(tok::l_brace)) {
1225         nextToken();
1226         parseBracedList();
1227         break;
1228       } else if (Style.Language == FormatStyle::LK_Java &&
1229                  FormatTok->is(Keywords.kw_interface)) {
1230         nextToken();
1231         break;
1232       }
1233       switch (FormatTok->Tok.getObjCKeywordID()) {
1234       case tok::objc_public:
1235       case tok::objc_protected:
1236       case tok::objc_package:
1237       case tok::objc_private:
1238         return parseAccessSpecifier();
1239       case tok::objc_interface:
1240       case tok::objc_implementation:
1241         return parseObjCInterfaceOrImplementation();
1242       case tok::objc_protocol:
1243         if (parseObjCProtocol())
1244           return;
1245         break;
1246       case tok::objc_end:
1247         return; // Handled by the caller.
1248       case tok::objc_optional:
1249       case tok::objc_required:
1250         nextToken();
1251         addUnwrappedLine();
1252         return;
1253       case tok::objc_autoreleasepool:
1254         nextToken();
1255         if (FormatTok->Tok.is(tok::l_brace)) {
1256           if (Style.BraceWrapping.AfterControlStatement ==
1257               FormatStyle::BWACS_Always)
1258             addUnwrappedLine();
1259           parseBlock(/*MustBeDeclaration=*/false);
1260         }
1261         addUnwrappedLine();
1262         return;
1263       case tok::objc_synchronized:
1264         nextToken();
1265         if (FormatTok->Tok.is(tok::l_paren))
1266           // Skip synchronization object
1267           parseParens();
1268         if (FormatTok->Tok.is(tok::l_brace)) {
1269           if (Style.BraceWrapping.AfterControlStatement ==
1270               FormatStyle::BWACS_Always)
1271             addUnwrappedLine();
1272           parseBlock(/*MustBeDeclaration=*/false);
1273         }
1274         addUnwrappedLine();
1275         return;
1276       case tok::objc_try:
1277         // This branch isn't strictly necessary (the kw_try case below would
1278         // do this too after the tok::at is parsed above).  But be explicit.
1279         parseTryCatch();
1280         return;
1281       default:
1282         break;
1283       }
1284       break;
1285     case tok::kw_concept:
1286       parseConcept();
1287       break;
1288     case tok::kw_requires:
1289       parseRequires();
1290       break;
1291     case tok::kw_enum:
1292       // Ignore if this is part of "template <enum ...".
1293       if (Previous && Previous->is(tok::less)) {
1294         nextToken();
1295         break;
1296       }
1297 
1298       // parseEnum falls through and does not yet add an unwrapped line as an
1299       // enum definition can start a structural element.
1300       if (!parseEnum())
1301         break;
1302       // This only applies for C++.
1303       if (!Style.isCpp()) {
1304         addUnwrappedLine();
1305         return;
1306       }
1307       break;
1308     case tok::kw_typedef:
1309       nextToken();
1310       if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS,
1311                              Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS,
1312                              Keywords.kw_CF_CLOSED_ENUM,
1313                              Keywords.kw_NS_CLOSED_ENUM))
1314         parseEnum();
1315       break;
1316     case tok::kw_struct:
1317     case tok::kw_union:
1318     case tok::kw_class:
1319       if (parseStructLike()) {
1320         return;
1321       }
1322       break;
1323     case tok::period:
1324       nextToken();
1325       // In Java, classes have an implicit static member "class".
1326       if (Style.Language == FormatStyle::LK_Java && FormatTok &&
1327           FormatTok->is(tok::kw_class))
1328         nextToken();
1329       if (Style.Language == FormatStyle::LK_JavaScript && FormatTok &&
1330           FormatTok->Tok.getIdentifierInfo())
1331         // JavaScript only has pseudo keywords, all keywords are allowed to
1332         // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6
1333         nextToken();
1334       break;
1335     case tok::semi:
1336       nextToken();
1337       addUnwrappedLine();
1338       return;
1339     case tok::r_brace:
1340       addUnwrappedLine();
1341       return;
1342     case tok::l_paren:
1343       parseParens();
1344       break;
1345     case tok::kw_operator:
1346       nextToken();
1347       if (FormatTok->isBinaryOperator())
1348         nextToken();
1349       break;
1350     case tok::caret:
1351       nextToken();
1352       if (FormatTok->Tok.isAnyIdentifier() ||
1353           FormatTok->isSimpleTypeSpecifier())
1354         nextToken();
1355       if (FormatTok->is(tok::l_paren))
1356         parseParens();
1357       if (FormatTok->is(tok::l_brace))
1358         parseChildBlock();
1359       break;
1360     case tok::l_brace:
1361       if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) {
1362         // A block outside of parentheses must be the last part of a
1363         // structural element.
1364         // FIXME: Figure out cases where this is not true, and add projections
1365         // for them (the one we know is missing are lambdas).
1366         if (Style.BraceWrapping.AfterFunction)
1367           addUnwrappedLine();
1368         FormatTok->setType(TT_FunctionLBrace);
1369         parseBlock(/*MustBeDeclaration=*/false);
1370         addUnwrappedLine();
1371         return;
1372       }
1373       // Otherwise this was a braced init list, and the structural
1374       // element continues.
1375       break;
1376     case tok::kw_try:
1377       if (Style.Language == FormatStyle::LK_JavaScript &&
1378           Line->MustBeDeclaration) {
1379         // field/method declaration.
1380         nextToken();
1381         break;
1382       }
1383       // We arrive here when parsing function-try blocks.
1384       if (Style.BraceWrapping.AfterFunction)
1385         addUnwrappedLine();
1386       parseTryCatch();
1387       return;
1388     case tok::identifier: {
1389       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) &&
1390           Line->MustBeDeclaration) {
1391         addUnwrappedLine();
1392         parseCSharpGenericTypeConstraint();
1393         break;
1394       }
1395       if (FormatTok->is(TT_MacroBlockEnd)) {
1396         addUnwrappedLine();
1397         return;
1398       }
1399 
1400       // Function declarations (as opposed to function expressions) are parsed
1401       // on their own unwrapped line by continuing this loop. Function
1402       // expressions (functions that are not on their own line) must not create
1403       // a new unwrapped line, so they are special cased below.
1404       size_t TokenCount = Line->Tokens.size();
1405       if (Style.Language == FormatStyle::LK_JavaScript &&
1406           FormatTok->is(Keywords.kw_function) &&
1407           (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is(
1408                                                      Keywords.kw_async)))) {
1409         tryToParseJSFunction();
1410         break;
1411       }
1412       if ((Style.Language == FormatStyle::LK_JavaScript ||
1413            Style.Language == FormatStyle::LK_Java) &&
1414           FormatTok->is(Keywords.kw_interface)) {
1415         if (Style.Language == FormatStyle::LK_JavaScript) {
1416           // In JavaScript/TypeScript, "interface" can be used as a standalone
1417           // identifier, e.g. in `var interface = 1;`. If "interface" is
1418           // followed by another identifier, it is very like to be an actual
1419           // interface declaration.
1420           unsigned StoredPosition = Tokens->getPosition();
1421           FormatToken *Next = Tokens->getNextToken();
1422           FormatTok = Tokens->setPosition(StoredPosition);
1423           if (Next && !mustBeJSIdent(Keywords, Next)) {
1424             nextToken();
1425             break;
1426           }
1427         }
1428         parseRecord();
1429         addUnwrappedLine();
1430         return;
1431       }
1432 
1433       if (FormatTok->is(Keywords.kw_interface)) {
1434         if (parseStructLike()) {
1435           return;
1436         }
1437         break;
1438       }
1439 
1440       if (Style.isCpp() && FormatTok->is(TT_StatementMacro)) {
1441         parseStatementMacro();
1442         return;
1443       }
1444 
1445       // See if the following token should start a new unwrapped line.
1446       StringRef Text = FormatTok->TokenText;
1447       nextToken();
1448 
1449       // JS doesn't have macros, and within classes colons indicate fields, not
1450       // labels.
1451       if (Style.Language == FormatStyle::LK_JavaScript)
1452         break;
1453 
1454       TokenCount = Line->Tokens.size();
1455       if (TokenCount == 1 ||
1456           (TokenCount == 2 && Line->Tokens.front().Tok->is(tok::comment))) {
1457         if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) {
1458           Line->Tokens.begin()->Tok->MustBreakBefore = true;
1459           parseLabel(!Style.IndentGotoLabels);
1460           return;
1461         }
1462         // Recognize function-like macro usages without trailing semicolon as
1463         // well as free-standing macros like Q_OBJECT.
1464         bool FunctionLike = FormatTok->is(tok::l_paren);
1465         if (FunctionLike)
1466           parseParens();
1467 
1468         bool FollowedByNewline =
1469             CommentsBeforeNextToken.empty()
1470                 ? FormatTok->NewlinesBefore > 0
1471                 : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
1472 
1473         if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) &&
1474             tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) {
1475           addUnwrappedLine();
1476           return;
1477         }
1478       }
1479       break;
1480     }
1481     case tok::equal:
1482       // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType
1483       // TT_FatArrow. The always start an expression or a child block if
1484       // followed by a curly.
1485       if (FormatTok->is(TT_FatArrow)) {
1486         nextToken();
1487         if (FormatTok->is(tok::l_brace)) {
1488           // C# may break after => if the next character is a newline.
1489           if (Style.isCSharp() && Style.BraceWrapping.AfterFunction == true) {
1490             // calling `addUnwrappedLine()` here causes odd parsing errors.
1491             FormatTok->MustBreakBefore = true;
1492           }
1493           parseChildBlock();
1494         }
1495         break;
1496       }
1497 
1498       nextToken();
1499       if (FormatTok->Tok.is(tok::l_brace)) {
1500         // Block kind should probably be set to BK_BracedInit for any language.
1501         // C# needs this change to ensure that array initialisers and object
1502         // initialisers are indented the same way.
1503         if (Style.isCSharp())
1504           FormatTok->setBlockKind(BK_BracedInit);
1505         nextToken();
1506         parseBracedList();
1507       } else if (Style.Language == FormatStyle::LK_Proto &&
1508                  FormatTok->Tok.is(tok::less)) {
1509         nextToken();
1510         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
1511                         /*ClosingBraceKind=*/tok::greater);
1512       }
1513       break;
1514     case tok::l_square:
1515       parseSquare();
1516       break;
1517     case tok::kw_new:
1518       parseNew();
1519       break;
1520     default:
1521       nextToken();
1522       break;
1523     }
1524   } while (!eof());
1525 }
1526 
tryToParsePropertyAccessor()1527 bool UnwrappedLineParser::tryToParsePropertyAccessor() {
1528   assert(FormatTok->is(tok::l_brace));
1529   if (!Style.isCSharp())
1530     return false;
1531   // See if it's a property accessor.
1532   if (FormatTok->Previous->isNot(tok::identifier))
1533     return false;
1534 
1535   // See if we are inside a property accessor.
1536   //
1537   // Record the current tokenPosition so that we can advance and
1538   // reset the current token. `Next` is not set yet so we need
1539   // another way to advance along the token stream.
1540   unsigned int StoredPosition = Tokens->getPosition();
1541   FormatToken *Tok = Tokens->getNextToken();
1542 
1543   // A trivial property accessor is of the form:
1544   // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set] }
1545   // Track these as they do not require line breaks to be introduced.
1546   bool HasGetOrSet = false;
1547   bool IsTrivialPropertyAccessor = true;
1548   while (!eof()) {
1549     if (Tok->isOneOf(tok::semi, tok::kw_public, tok::kw_private,
1550                      tok::kw_protected, Keywords.kw_internal, Keywords.kw_get,
1551                      Keywords.kw_set)) {
1552       if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_set))
1553         HasGetOrSet = true;
1554       Tok = Tokens->getNextToken();
1555       continue;
1556     }
1557     if (Tok->isNot(tok::r_brace))
1558       IsTrivialPropertyAccessor = false;
1559     break;
1560   }
1561 
1562   if (!HasGetOrSet) {
1563     Tokens->setPosition(StoredPosition);
1564     return false;
1565   }
1566 
1567   // Try to parse the property accessor:
1568   // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties
1569   Tokens->setPosition(StoredPosition);
1570   if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction == true)
1571     addUnwrappedLine();
1572   nextToken();
1573   do {
1574     switch (FormatTok->Tok.getKind()) {
1575     case tok::r_brace:
1576       nextToken();
1577       if (FormatTok->is(tok::equal)) {
1578         while (!eof() && FormatTok->isNot(tok::semi))
1579           nextToken();
1580         nextToken();
1581       }
1582       addUnwrappedLine();
1583       return true;
1584     case tok::l_brace:
1585       ++Line->Level;
1586       parseBlock(/*MustBeDeclaration=*/true);
1587       addUnwrappedLine();
1588       --Line->Level;
1589       break;
1590     case tok::equal:
1591       if (FormatTok->is(TT_FatArrow)) {
1592         ++Line->Level;
1593         do {
1594           nextToken();
1595         } while (!eof() && FormatTok->isNot(tok::semi));
1596         nextToken();
1597         addUnwrappedLine();
1598         --Line->Level;
1599         break;
1600       }
1601       nextToken();
1602       break;
1603     default:
1604       if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_set) &&
1605           !IsTrivialPropertyAccessor) {
1606         // Non-trivial get/set needs to be on its own line.
1607         addUnwrappedLine();
1608       }
1609       nextToken();
1610     }
1611   } while (!eof());
1612 
1613   // Unreachable for well-formed code (paired '{' and '}').
1614   return true;
1615 }
1616 
tryToParseLambda()1617 bool UnwrappedLineParser::tryToParseLambda() {
1618   if (!Style.isCpp()) {
1619     nextToken();
1620     return false;
1621   }
1622   assert(FormatTok->is(tok::l_square));
1623   FormatToken &LSquare = *FormatTok;
1624   if (!tryToParseLambdaIntroducer())
1625     return false;
1626 
1627   bool SeenArrow = false;
1628 
1629   while (FormatTok->isNot(tok::l_brace)) {
1630     if (FormatTok->isSimpleTypeSpecifier()) {
1631       nextToken();
1632       continue;
1633     }
1634     switch (FormatTok->Tok.getKind()) {
1635     case tok::l_brace:
1636       break;
1637     case tok::l_paren:
1638       parseParens();
1639       break;
1640     case tok::amp:
1641     case tok::star:
1642     case tok::kw_const:
1643     case tok::comma:
1644     case tok::less:
1645     case tok::greater:
1646     case tok::identifier:
1647     case tok::numeric_constant:
1648     case tok::coloncolon:
1649     case tok::kw_class:
1650     case tok::kw_mutable:
1651     case tok::kw_noexcept:
1652     case tok::kw_template:
1653     case tok::kw_typename:
1654       nextToken();
1655       break;
1656     // Specialization of a template with an integer parameter can contain
1657     // arithmetic, logical, comparison and ternary operators.
1658     //
1659     // FIXME: This also accepts sequences of operators that are not in the scope
1660     // of a template argument list.
1661     //
1662     // In a C++ lambda a template type can only occur after an arrow. We use
1663     // this as an heuristic to distinguish between Objective-C expressions
1664     // followed by an `a->b` expression, such as:
1665     // ([obj func:arg] + a->b)
1666     // Otherwise the code below would parse as a lambda.
1667     //
1668     // FIXME: This heuristic is incorrect for C++20 generic lambdas with
1669     // explicit template lists: []<bool b = true && false>(U &&u){}
1670     case tok::plus:
1671     case tok::minus:
1672     case tok::exclaim:
1673     case tok::tilde:
1674     case tok::slash:
1675     case tok::percent:
1676     case tok::lessless:
1677     case tok::pipe:
1678     case tok::pipepipe:
1679     case tok::ampamp:
1680     case tok::caret:
1681     case tok::equalequal:
1682     case tok::exclaimequal:
1683     case tok::greaterequal:
1684     case tok::lessequal:
1685     case tok::question:
1686     case tok::colon:
1687     case tok::ellipsis:
1688     case tok::kw_true:
1689     case tok::kw_false:
1690       if (SeenArrow) {
1691         nextToken();
1692         break;
1693       }
1694       return true;
1695     case tok::arrow:
1696       // This might or might not actually be a lambda arrow (this could be an
1697       // ObjC method invocation followed by a dereferencing arrow). We might
1698       // reset this back to TT_Unknown in TokenAnnotator.
1699       FormatTok->setType(TT_LambdaArrow);
1700       SeenArrow = true;
1701       nextToken();
1702       break;
1703     default:
1704       return true;
1705     }
1706   }
1707   FormatTok->setType(TT_LambdaLBrace);
1708   LSquare.setType(TT_LambdaLSquare);
1709   parseChildBlock();
1710   return true;
1711 }
1712 
tryToParseLambdaIntroducer()1713 bool UnwrappedLineParser::tryToParseLambdaIntroducer() {
1714   const FormatToken *Previous = FormatTok->Previous;
1715   if (Previous &&
1716       (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new,
1717                          tok::kw_delete, tok::l_square) ||
1718        FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() ||
1719        Previous->isSimpleTypeSpecifier())) {
1720     nextToken();
1721     return false;
1722   }
1723   nextToken();
1724   if (FormatTok->is(tok::l_square)) {
1725     return false;
1726   }
1727   parseSquare(/*LambdaIntroducer=*/true);
1728   return true;
1729 }
1730 
tryToParseJSFunction()1731 void UnwrappedLineParser::tryToParseJSFunction() {
1732   assert(FormatTok->is(Keywords.kw_function) ||
1733          FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function));
1734   if (FormatTok->is(Keywords.kw_async))
1735     nextToken();
1736   // Consume "function".
1737   nextToken();
1738 
1739   // Consume * (generator function). Treat it like C++'s overloaded operators.
1740   if (FormatTok->is(tok::star)) {
1741     FormatTok->setType(TT_OverloadedOperator);
1742     nextToken();
1743   }
1744 
1745   // Consume function name.
1746   if (FormatTok->is(tok::identifier))
1747     nextToken();
1748 
1749   if (FormatTok->isNot(tok::l_paren))
1750     return;
1751 
1752   // Parse formal parameter list.
1753   parseParens();
1754 
1755   if (FormatTok->is(tok::colon)) {
1756     // Parse a type definition.
1757     nextToken();
1758 
1759     // Eat the type declaration. For braced inline object types, balance braces,
1760     // otherwise just parse until finding an l_brace for the function body.
1761     if (FormatTok->is(tok::l_brace))
1762       tryToParseBracedList();
1763     else
1764       while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof())
1765         nextToken();
1766   }
1767 
1768   if (FormatTok->is(tok::semi))
1769     return;
1770 
1771   parseChildBlock();
1772 }
1773 
tryToParseBracedList()1774 bool UnwrappedLineParser::tryToParseBracedList() {
1775   if (FormatTok->is(BK_Unknown))
1776     calculateBraceTypes();
1777   assert(FormatTok->isNot(BK_Unknown));
1778   if (FormatTok->is(BK_Block))
1779     return false;
1780   nextToken();
1781   parseBracedList();
1782   return true;
1783 }
1784 
parseBracedList(bool ContinueOnSemicolons,bool IsEnum,tok::TokenKind ClosingBraceKind)1785 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons,
1786                                           bool IsEnum,
1787                                           tok::TokenKind ClosingBraceKind) {
1788   bool HasError = false;
1789 
1790   // FIXME: Once we have an expression parser in the UnwrappedLineParser,
1791   // replace this by using parseAssigmentExpression() inside.
1792   do {
1793     if (Style.isCSharp()) {
1794       if (FormatTok->is(TT_FatArrow)) {
1795         nextToken();
1796         // Fat arrows can be followed by simple expressions or by child blocks
1797         // in curly braces.
1798         if (FormatTok->is(tok::l_brace)) {
1799           parseChildBlock();
1800           continue;
1801         }
1802       }
1803     }
1804     if (Style.Language == FormatStyle::LK_JavaScript) {
1805       if (FormatTok->is(Keywords.kw_function) ||
1806           FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) {
1807         tryToParseJSFunction();
1808         continue;
1809       }
1810       if (FormatTok->is(TT_FatArrow)) {
1811         nextToken();
1812         // Fat arrows can be followed by simple expressions or by child blocks
1813         // in curly braces.
1814         if (FormatTok->is(tok::l_brace)) {
1815           parseChildBlock();
1816           continue;
1817         }
1818       }
1819       if (FormatTok->is(tok::l_brace)) {
1820         // Could be a method inside of a braced list `{a() { return 1; }}`.
1821         if (tryToParseBracedList())
1822           continue;
1823         parseChildBlock();
1824       }
1825     }
1826     if (FormatTok->Tok.getKind() == ClosingBraceKind) {
1827       if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
1828         addUnwrappedLine();
1829       nextToken();
1830       return !HasError;
1831     }
1832     switch (FormatTok->Tok.getKind()) {
1833     case tok::caret:
1834       nextToken();
1835       if (FormatTok->is(tok::l_brace)) {
1836         parseChildBlock();
1837       }
1838       break;
1839     case tok::l_square:
1840       if (Style.isCSharp())
1841         parseSquare();
1842       else
1843         tryToParseLambda();
1844       break;
1845     case tok::l_paren:
1846       parseParens();
1847       // JavaScript can just have free standing methods and getters/setters in
1848       // object literals. Detect them by a "{" following ")".
1849       if (Style.Language == FormatStyle::LK_JavaScript) {
1850         if (FormatTok->is(tok::l_brace))
1851           parseChildBlock();
1852         break;
1853       }
1854       break;
1855     case tok::l_brace:
1856       // Assume there are no blocks inside a braced init list apart
1857       // from the ones we explicitly parse out (like lambdas).
1858       FormatTok->setBlockKind(BK_BracedInit);
1859       nextToken();
1860       parseBracedList();
1861       break;
1862     case tok::less:
1863       if (Style.Language == FormatStyle::LK_Proto) {
1864         nextToken();
1865         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
1866                         /*ClosingBraceKind=*/tok::greater);
1867       } else {
1868         nextToken();
1869       }
1870       break;
1871     case tok::semi:
1872       // JavaScript (or more precisely TypeScript) can have semicolons in braced
1873       // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be
1874       // used for error recovery if we have otherwise determined that this is
1875       // a braced list.
1876       if (Style.Language == FormatStyle::LK_JavaScript) {
1877         nextToken();
1878         break;
1879       }
1880       HasError = true;
1881       if (!ContinueOnSemicolons)
1882         return !HasError;
1883       nextToken();
1884       break;
1885     case tok::comma:
1886       nextToken();
1887       if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
1888         addUnwrappedLine();
1889       break;
1890     default:
1891       nextToken();
1892       break;
1893     }
1894   } while (!eof());
1895   return false;
1896 }
1897 
parseParens()1898 void UnwrappedLineParser::parseParens() {
1899   assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected.");
1900   nextToken();
1901   do {
1902     switch (FormatTok->Tok.getKind()) {
1903     case tok::l_paren:
1904       parseParens();
1905       if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace))
1906         parseChildBlock();
1907       break;
1908     case tok::r_paren:
1909       nextToken();
1910       return;
1911     case tok::r_brace:
1912       // A "}" inside parenthesis is an error if there wasn't a matching "{".
1913       return;
1914     case tok::l_square:
1915       tryToParseLambda();
1916       break;
1917     case tok::l_brace:
1918       if (!tryToParseBracedList())
1919         parseChildBlock();
1920       break;
1921     case tok::at:
1922       nextToken();
1923       if (FormatTok->Tok.is(tok::l_brace)) {
1924         nextToken();
1925         parseBracedList();
1926       }
1927       break;
1928     case tok::kw_class:
1929       if (Style.Language == FormatStyle::LK_JavaScript)
1930         parseRecord(/*ParseAsExpr=*/true);
1931       else
1932         nextToken();
1933       break;
1934     case tok::identifier:
1935       if (Style.Language == FormatStyle::LK_JavaScript &&
1936           (FormatTok->is(Keywords.kw_function) ||
1937            FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)))
1938         tryToParseJSFunction();
1939       else
1940         nextToken();
1941       break;
1942     default:
1943       nextToken();
1944       break;
1945     }
1946   } while (!eof());
1947 }
1948 
parseSquare(bool LambdaIntroducer)1949 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) {
1950   if (!LambdaIntroducer) {
1951     assert(FormatTok->Tok.is(tok::l_square) && "'[' expected.");
1952     if (tryToParseLambda())
1953       return;
1954   }
1955   do {
1956     switch (FormatTok->Tok.getKind()) {
1957     case tok::l_paren:
1958       parseParens();
1959       break;
1960     case tok::r_square:
1961       nextToken();
1962       return;
1963     case tok::r_brace:
1964       // A "}" inside parenthesis is an error if there wasn't a matching "{".
1965       return;
1966     case tok::l_square:
1967       parseSquare();
1968       break;
1969     case tok::l_brace: {
1970       if (!tryToParseBracedList())
1971         parseChildBlock();
1972       break;
1973     }
1974     case tok::at:
1975       nextToken();
1976       if (FormatTok->Tok.is(tok::l_brace)) {
1977         nextToken();
1978         parseBracedList();
1979       }
1980       break;
1981     default:
1982       nextToken();
1983       break;
1984     }
1985   } while (!eof());
1986 }
1987 
parseIfThenElse()1988 void UnwrappedLineParser::parseIfThenElse() {
1989   assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected");
1990   nextToken();
1991   if (FormatTok->Tok.isOneOf(tok::kw_constexpr, tok::identifier))
1992     nextToken();
1993   if (FormatTok->Tok.is(tok::l_paren))
1994     parseParens();
1995   // handle [[likely]] / [[unlikely]]
1996   if (FormatTok->is(tok::l_square) && tryToParseSimpleAttribute())
1997     parseSquare();
1998   bool NeedsUnwrappedLine = false;
1999   if (FormatTok->Tok.is(tok::l_brace)) {
2000     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2001     parseBlock(/*MustBeDeclaration=*/false);
2002     if (Style.BraceWrapping.BeforeElse)
2003       addUnwrappedLine();
2004     else
2005       NeedsUnwrappedLine = true;
2006   } else {
2007     addUnwrappedLine();
2008     ++Line->Level;
2009     parseStructuralElement();
2010     --Line->Level;
2011   }
2012   if (FormatTok->Tok.is(tok::kw_else)) {
2013     nextToken();
2014     // handle [[likely]] / [[unlikely]]
2015     if (FormatTok->Tok.is(tok::l_square) && tryToParseSimpleAttribute())
2016       parseSquare();
2017     if (FormatTok->Tok.is(tok::l_brace)) {
2018       CompoundStatementIndenter Indenter(this, Style, Line->Level);
2019       parseBlock(/*MustBeDeclaration=*/false);
2020       addUnwrappedLine();
2021     } else if (FormatTok->Tok.is(tok::kw_if)) {
2022       parseIfThenElse();
2023     } else {
2024       addUnwrappedLine();
2025       ++Line->Level;
2026       parseStructuralElement();
2027       if (FormatTok->is(tok::eof))
2028         addUnwrappedLine();
2029       --Line->Level;
2030     }
2031   } else if (NeedsUnwrappedLine) {
2032     addUnwrappedLine();
2033   }
2034 }
2035 
parseTryCatch()2036 void UnwrappedLineParser::parseTryCatch() {
2037   assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
2038   nextToken();
2039   bool NeedsUnwrappedLine = false;
2040   if (FormatTok->is(tok::colon)) {
2041     // We are in a function try block, what comes is an initializer list.
2042     nextToken();
2043 
2044     // In case identifiers were removed by clang-tidy, what might follow is
2045     // multiple commas in sequence - before the first identifier.
2046     while (FormatTok->is(tok::comma))
2047       nextToken();
2048 
2049     while (FormatTok->is(tok::identifier)) {
2050       nextToken();
2051       if (FormatTok->is(tok::l_paren))
2052         parseParens();
2053       if (FormatTok->Previous && FormatTok->Previous->is(tok::identifier) &&
2054           FormatTok->is(tok::l_brace)) {
2055         do {
2056           nextToken();
2057         } while (!FormatTok->is(tok::r_brace));
2058         nextToken();
2059       }
2060 
2061       // In case identifiers were removed by clang-tidy, what might follow is
2062       // multiple commas in sequence - after the first identifier.
2063       while (FormatTok->is(tok::comma))
2064         nextToken();
2065     }
2066   }
2067   // Parse try with resource.
2068   if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) {
2069     parseParens();
2070   }
2071   if (FormatTok->is(tok::l_brace)) {
2072     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2073     parseBlock(/*MustBeDeclaration=*/false);
2074     if (Style.BraceWrapping.BeforeCatch) {
2075       addUnwrappedLine();
2076     } else {
2077       NeedsUnwrappedLine = true;
2078     }
2079   } else if (!FormatTok->is(tok::kw_catch)) {
2080     // The C++ standard requires a compound-statement after a try.
2081     // If there's none, we try to assume there's a structuralElement
2082     // and try to continue.
2083     addUnwrappedLine();
2084     ++Line->Level;
2085     parseStructuralElement();
2086     --Line->Level;
2087   }
2088   while (1) {
2089     if (FormatTok->is(tok::at))
2090       nextToken();
2091     if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except,
2092                              tok::kw___finally) ||
2093           ((Style.Language == FormatStyle::LK_Java ||
2094             Style.Language == FormatStyle::LK_JavaScript) &&
2095            FormatTok->is(Keywords.kw_finally)) ||
2096           (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) ||
2097            FormatTok->Tok.isObjCAtKeyword(tok::objc_finally))))
2098       break;
2099     nextToken();
2100     while (FormatTok->isNot(tok::l_brace)) {
2101       if (FormatTok->is(tok::l_paren)) {
2102         parseParens();
2103         continue;
2104       }
2105       if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof))
2106         return;
2107       nextToken();
2108     }
2109     NeedsUnwrappedLine = false;
2110     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2111     parseBlock(/*MustBeDeclaration=*/false);
2112     if (Style.BraceWrapping.BeforeCatch)
2113       addUnwrappedLine();
2114     else
2115       NeedsUnwrappedLine = true;
2116   }
2117   if (NeedsUnwrappedLine)
2118     addUnwrappedLine();
2119 }
2120 
parseNamespace()2121 void UnwrappedLineParser::parseNamespace() {
2122   assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) &&
2123          "'namespace' expected");
2124 
2125   const FormatToken &InitialToken = *FormatTok;
2126   nextToken();
2127   if (InitialToken.is(TT_NamespaceMacro)) {
2128     parseParens();
2129   } else {
2130     while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline,
2131                               tok::l_square)) {
2132       if (FormatTok->is(tok::l_square))
2133         parseSquare();
2134       else
2135         nextToken();
2136     }
2137   }
2138   if (FormatTok->Tok.is(tok::l_brace)) {
2139     if (ShouldBreakBeforeBrace(Style, InitialToken))
2140       addUnwrappedLine();
2141 
2142     unsigned AddLevels =
2143         Style.NamespaceIndentation == FormatStyle::NI_All ||
2144                 (Style.NamespaceIndentation == FormatStyle::NI_Inner &&
2145                  DeclarationScopeStack.size() > 1)
2146             ? 1u
2147             : 0u;
2148     bool ManageWhitesmithsBraces =
2149         AddLevels == 0u &&
2150         Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
2151 
2152     // If we're in Whitesmiths mode, indent the brace if we're not indenting
2153     // the whole block.
2154     if (ManageWhitesmithsBraces)
2155       ++Line->Level;
2156 
2157     parseBlock(/*MustBeDeclaration=*/true, AddLevels,
2158                /*MunchSemi=*/true,
2159                /*UnindentWhitesmithsBraces=*/ManageWhitesmithsBraces);
2160 
2161     // Munch the semicolon after a namespace. This is more common than one would
2162     // think. Putting the semicolon into its own line is very ugly.
2163     if (FormatTok->Tok.is(tok::semi))
2164       nextToken();
2165 
2166     addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep);
2167 
2168     if (ManageWhitesmithsBraces)
2169       --Line->Level;
2170   }
2171   // FIXME: Add error handling.
2172 }
2173 
parseNew()2174 void UnwrappedLineParser::parseNew() {
2175   assert(FormatTok->is(tok::kw_new) && "'new' expected");
2176   nextToken();
2177 
2178   if (Style.isCSharp()) {
2179     do {
2180       if (FormatTok->is(tok::l_brace))
2181         parseBracedList();
2182 
2183       if (FormatTok->isOneOf(tok::semi, tok::comma))
2184         return;
2185 
2186       nextToken();
2187     } while (!eof());
2188   }
2189 
2190   if (Style.Language != FormatStyle::LK_Java)
2191     return;
2192 
2193   // In Java, we can parse everything up to the parens, which aren't optional.
2194   do {
2195     // There should not be a ;, { or } before the new's open paren.
2196     if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace))
2197       return;
2198 
2199     // Consume the parens.
2200     if (FormatTok->is(tok::l_paren)) {
2201       parseParens();
2202 
2203       // If there is a class body of an anonymous class, consume that as child.
2204       if (FormatTok->is(tok::l_brace))
2205         parseChildBlock();
2206       return;
2207     }
2208     nextToken();
2209   } while (!eof());
2210 }
2211 
parseForOrWhileLoop()2212 void UnwrappedLineParser::parseForOrWhileLoop() {
2213   assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) &&
2214          "'for', 'while' or foreach macro expected");
2215   nextToken();
2216   // JS' for await ( ...
2217   if (Style.Language == FormatStyle::LK_JavaScript &&
2218       FormatTok->is(Keywords.kw_await))
2219     nextToken();
2220   if (FormatTok->Tok.is(tok::l_paren))
2221     parseParens();
2222   if (FormatTok->Tok.is(tok::l_brace)) {
2223     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2224     parseBlock(/*MustBeDeclaration=*/false);
2225     addUnwrappedLine();
2226   } else {
2227     addUnwrappedLine();
2228     ++Line->Level;
2229     parseStructuralElement();
2230     --Line->Level;
2231   }
2232 }
2233 
parseDoWhile()2234 void UnwrappedLineParser::parseDoWhile() {
2235   assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected");
2236   nextToken();
2237   if (FormatTok->Tok.is(tok::l_brace)) {
2238     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2239     parseBlock(/*MustBeDeclaration=*/false);
2240     if (Style.BraceWrapping.BeforeWhile)
2241       addUnwrappedLine();
2242   } else {
2243     addUnwrappedLine();
2244     ++Line->Level;
2245     parseStructuralElement();
2246     --Line->Level;
2247   }
2248 
2249   // FIXME: Add error handling.
2250   if (!FormatTok->Tok.is(tok::kw_while)) {
2251     addUnwrappedLine();
2252     return;
2253   }
2254 
2255   // If in Whitesmiths mode, the line with the while() needs to be indented
2256   // to the same level as the block.
2257   if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
2258     ++Line->Level;
2259 
2260   nextToken();
2261   parseStructuralElement();
2262 }
2263 
parseLabel(bool LeftAlignLabel)2264 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) {
2265   nextToken();
2266   unsigned OldLineLevel = Line->Level;
2267   if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0))
2268     --Line->Level;
2269   if (LeftAlignLabel)
2270     Line->Level = 0;
2271 
2272   if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() &&
2273       FormatTok->Tok.is(tok::l_brace)) {
2274 
2275     CompoundStatementIndenter Indenter(this, Line->Level,
2276                                        Style.BraceWrapping.AfterCaseLabel,
2277                                        Style.BraceWrapping.IndentBraces);
2278     parseBlock(/*MustBeDeclaration=*/false);
2279     if (FormatTok->Tok.is(tok::kw_break)) {
2280       if (Style.BraceWrapping.AfterControlStatement ==
2281           FormatStyle::BWACS_Always) {
2282         addUnwrappedLine();
2283         if (!Style.IndentCaseBlocks &&
2284             Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
2285           Line->Level++;
2286         }
2287       }
2288       parseStructuralElement();
2289     }
2290     addUnwrappedLine();
2291   } else {
2292     if (FormatTok->is(tok::semi))
2293       nextToken();
2294     addUnwrappedLine();
2295   }
2296   Line->Level = OldLineLevel;
2297   if (FormatTok->isNot(tok::l_brace)) {
2298     parseStructuralElement();
2299     addUnwrappedLine();
2300   }
2301 }
2302 
parseCaseLabel()2303 void UnwrappedLineParser::parseCaseLabel() {
2304   assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected");
2305 
2306   // FIXME: fix handling of complex expressions here.
2307   do {
2308     nextToken();
2309   } while (!eof() && !FormatTok->Tok.is(tok::colon));
2310   parseLabel();
2311 }
2312 
parseSwitch()2313 void UnwrappedLineParser::parseSwitch() {
2314   assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected");
2315   nextToken();
2316   if (FormatTok->Tok.is(tok::l_paren))
2317     parseParens();
2318   if (FormatTok->Tok.is(tok::l_brace)) {
2319     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2320     parseBlock(/*MustBeDeclaration=*/false);
2321     addUnwrappedLine();
2322   } else {
2323     addUnwrappedLine();
2324     ++Line->Level;
2325     parseStructuralElement();
2326     --Line->Level;
2327   }
2328 }
2329 
parseAccessSpecifier()2330 void UnwrappedLineParser::parseAccessSpecifier() {
2331   nextToken();
2332   // Understand Qt's slots.
2333   if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots))
2334     nextToken();
2335   // Otherwise, we don't know what it is, and we'd better keep the next token.
2336   if (FormatTok->Tok.is(tok::colon))
2337     nextToken();
2338   addUnwrappedLine();
2339 }
2340 
parseConcept()2341 void UnwrappedLineParser::parseConcept() {
2342   assert(FormatTok->Tok.is(tok::kw_concept) && "'concept' expected");
2343   nextToken();
2344   if (!FormatTok->Tok.is(tok::identifier))
2345     return;
2346   nextToken();
2347   if (!FormatTok->Tok.is(tok::equal))
2348     return;
2349   nextToken();
2350   if (FormatTok->Tok.is(tok::kw_requires)) {
2351     nextToken();
2352     parseRequiresExpression(Line->Level);
2353   } else {
2354     parseConstraintExpression(Line->Level);
2355   }
2356 }
2357 
parseRequiresExpression(unsigned int OriginalLevel)2358 void UnwrappedLineParser::parseRequiresExpression(unsigned int OriginalLevel) {
2359   // requires (R range)
2360   if (FormatTok->Tok.is(tok::l_paren)) {
2361     parseParens();
2362     if (Style.IndentRequires && OriginalLevel != Line->Level) {
2363       addUnwrappedLine();
2364       --Line->Level;
2365     }
2366   }
2367 
2368   if (FormatTok->Tok.is(tok::l_brace)) {
2369     if (Style.BraceWrapping.AfterFunction)
2370       addUnwrappedLine();
2371     FormatTok->setType(TT_FunctionLBrace);
2372     parseBlock(/*MustBeDeclaration=*/false);
2373     addUnwrappedLine();
2374   } else {
2375     parseConstraintExpression(OriginalLevel);
2376   }
2377 }
2378 
parseConstraintExpression(unsigned int OriginalLevel)2379 void UnwrappedLineParser::parseConstraintExpression(
2380     unsigned int OriginalLevel) {
2381   // requires Id<T> && Id<T> || Id<T>
2382   while (
2383       FormatTok->isOneOf(tok::identifier, tok::kw_requires, tok::coloncolon)) {
2384     nextToken();
2385     while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::less,
2386                               tok::greater, tok::comma, tok::ellipsis)) {
2387       if (FormatTok->Tok.is(tok::less)) {
2388         parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
2389                         /*ClosingBraceKind=*/tok::greater);
2390         continue;
2391       }
2392       nextToken();
2393     }
2394     if (FormatTok->Tok.is(tok::kw_requires)) {
2395       parseRequiresExpression(OriginalLevel);
2396     }
2397     if (FormatTok->Tok.is(tok::less)) {
2398       parseBracedList(/*ContinueOnSemicolons=*/false, /*IsEnum=*/false,
2399                       /*ClosingBraceKind=*/tok::greater);
2400     }
2401 
2402     if (FormatTok->Tok.is(tok::l_paren)) {
2403       parseParens();
2404     }
2405     if (FormatTok->Tok.is(tok::l_brace)) {
2406       if (Style.BraceWrapping.AfterFunction)
2407         addUnwrappedLine();
2408       FormatTok->setType(TT_FunctionLBrace);
2409       parseBlock(/*MustBeDeclaration=*/false);
2410     }
2411     if (FormatTok->Tok.is(tok::semi)) {
2412       // Eat any trailing semi.
2413       nextToken();
2414       addUnwrappedLine();
2415     }
2416     if (FormatTok->Tok.is(tok::colon)) {
2417       return;
2418     }
2419     if (!FormatTok->Tok.isOneOf(tok::ampamp, tok::pipepipe)) {
2420       if (FormatTok->Previous &&
2421           !FormatTok->Previous->isOneOf(tok::identifier, tok::kw_requires,
2422                                         tok::coloncolon)) {
2423         addUnwrappedLine();
2424       }
2425       if (Style.IndentRequires && OriginalLevel != Line->Level) {
2426         --Line->Level;
2427       }
2428       break;
2429     } else {
2430       FormatTok->setType(TT_ConstraintJunctions);
2431     }
2432 
2433     nextToken();
2434   }
2435 }
2436 
parseRequires()2437 void UnwrappedLineParser::parseRequires() {
2438   assert(FormatTok->Tok.is(tok::kw_requires) && "'requires' expected");
2439 
2440   unsigned OriginalLevel = Line->Level;
2441   if (FormatTok->Previous && FormatTok->Previous->is(tok::greater)) {
2442     addUnwrappedLine();
2443     if (Style.IndentRequires) {
2444       Line->Level++;
2445     }
2446   }
2447   nextToken();
2448 
2449   parseRequiresExpression(OriginalLevel);
2450 }
2451 
parseEnum()2452 bool UnwrappedLineParser::parseEnum() {
2453   // Won't be 'enum' for NS_ENUMs.
2454   if (FormatTok->Tok.is(tok::kw_enum))
2455     nextToken();
2456 
2457   // In TypeScript, "enum" can also be used as property name, e.g. in interface
2458   // declarations. An "enum" keyword followed by a colon would be a syntax
2459   // error and thus assume it is just an identifier.
2460   if (Style.Language == FormatStyle::LK_JavaScript &&
2461       FormatTok->isOneOf(tok::colon, tok::question))
2462     return false;
2463 
2464   // In protobuf, "enum" can be used as a field name.
2465   if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal))
2466     return false;
2467 
2468   // Eat up enum class ...
2469   if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct))
2470     nextToken();
2471 
2472   while (FormatTok->Tok.getIdentifierInfo() ||
2473          FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less,
2474                             tok::greater, tok::comma, tok::question)) {
2475     nextToken();
2476     // We can have macros or attributes in between 'enum' and the enum name.
2477     if (FormatTok->is(tok::l_paren))
2478       parseParens();
2479     if (FormatTok->is(tok::identifier)) {
2480       nextToken();
2481       // If there are two identifiers in a row, this is likely an elaborate
2482       // return type. In Java, this can be "implements", etc.
2483       if (Style.isCpp() && FormatTok->is(tok::identifier))
2484         return false;
2485     }
2486   }
2487 
2488   // Just a declaration or something is wrong.
2489   if (FormatTok->isNot(tok::l_brace))
2490     return true;
2491   FormatTok->setBlockKind(BK_Block);
2492 
2493   if (Style.Language == FormatStyle::LK_Java) {
2494     // Java enums are different.
2495     parseJavaEnumBody();
2496     return true;
2497   }
2498   if (Style.Language == FormatStyle::LK_Proto) {
2499     parseBlock(/*MustBeDeclaration=*/true);
2500     return true;
2501   }
2502 
2503   if (!Style.AllowShortEnumsOnASingleLine)
2504     addUnwrappedLine();
2505   // Parse enum body.
2506   nextToken();
2507   if (!Style.AllowShortEnumsOnASingleLine) {
2508     addUnwrappedLine();
2509     Line->Level += 1;
2510   }
2511   bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true,
2512                                    /*IsEnum=*/true);
2513   if (!Style.AllowShortEnumsOnASingleLine)
2514     Line->Level -= 1;
2515   if (HasError) {
2516     if (FormatTok->is(tok::semi))
2517       nextToken();
2518     addUnwrappedLine();
2519   }
2520   return true;
2521 
2522   // There is no addUnwrappedLine() here so that we fall through to parsing a
2523   // structural element afterwards. Thus, in "enum A {} n, m;",
2524   // "} n, m;" will end up in one unwrapped line.
2525 }
2526 
parseStructLike()2527 bool UnwrappedLineParser::parseStructLike() {
2528   // parseRecord falls through and does not yet add an unwrapped line as a
2529   // record declaration or definition can start a structural element.
2530   parseRecord();
2531   // This does not apply to Java, JavaScript and C#.
2532   if (Style.Language == FormatStyle::LK_Java ||
2533       Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) {
2534     if (FormatTok->is(tok::semi))
2535       nextToken();
2536     addUnwrappedLine();
2537     return true;
2538   }
2539   return false;
2540 }
2541 
2542 namespace {
2543 // A class used to set and restore the Token position when peeking
2544 // ahead in the token source.
2545 class ScopedTokenPosition {
2546   unsigned StoredPosition;
2547   FormatTokenSource *Tokens;
2548 
2549 public:
ScopedTokenPosition(FormatTokenSource * Tokens)2550   ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) {
2551     assert(Tokens && "Tokens expected to not be null");
2552     StoredPosition = Tokens->getPosition();
2553   }
2554 
~ScopedTokenPosition()2555   ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); }
2556 };
2557 } // namespace
2558 
2559 // Look to see if we have [[ by looking ahead, if
2560 // its not then rewind to the original position.
tryToParseSimpleAttribute()2561 bool UnwrappedLineParser::tryToParseSimpleAttribute() {
2562   ScopedTokenPosition AutoPosition(Tokens);
2563   FormatToken *Tok = Tokens->getNextToken();
2564   // We already read the first [ check for the second.
2565   if (Tok && !Tok->is(tok::l_square)) {
2566     return false;
2567   }
2568   // Double check that the attribute is just something
2569   // fairly simple.
2570   while (Tok) {
2571     if (Tok->is(tok::r_square)) {
2572       break;
2573     }
2574     Tok = Tokens->getNextToken();
2575   }
2576   Tok = Tokens->getNextToken();
2577   if (Tok && !Tok->is(tok::r_square)) {
2578     return false;
2579   }
2580   Tok = Tokens->getNextToken();
2581   if (Tok && Tok->is(tok::semi)) {
2582     return false;
2583   }
2584   return true;
2585 }
2586 
parseJavaEnumBody()2587 void UnwrappedLineParser::parseJavaEnumBody() {
2588   // Determine whether the enum is simple, i.e. does not have a semicolon or
2589   // constants with class bodies. Simple enums can be formatted like braced
2590   // lists, contracted to a single line, etc.
2591   unsigned StoredPosition = Tokens->getPosition();
2592   bool IsSimple = true;
2593   FormatToken *Tok = Tokens->getNextToken();
2594   while (Tok) {
2595     if (Tok->is(tok::r_brace))
2596       break;
2597     if (Tok->isOneOf(tok::l_brace, tok::semi)) {
2598       IsSimple = false;
2599       break;
2600     }
2601     // FIXME: This will also mark enums with braces in the arguments to enum
2602     // constants as "not simple". This is probably fine in practice, though.
2603     Tok = Tokens->getNextToken();
2604   }
2605   FormatTok = Tokens->setPosition(StoredPosition);
2606 
2607   if (IsSimple) {
2608     nextToken();
2609     parseBracedList();
2610     addUnwrappedLine();
2611     return;
2612   }
2613 
2614   // Parse the body of a more complex enum.
2615   // First add a line for everything up to the "{".
2616   nextToken();
2617   addUnwrappedLine();
2618   ++Line->Level;
2619 
2620   // Parse the enum constants.
2621   while (FormatTok) {
2622     if (FormatTok->is(tok::l_brace)) {
2623       // Parse the constant's class body.
2624       parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u,
2625                  /*MunchSemi=*/false);
2626     } else if (FormatTok->is(tok::l_paren)) {
2627       parseParens();
2628     } else if (FormatTok->is(tok::comma)) {
2629       nextToken();
2630       addUnwrappedLine();
2631     } else if (FormatTok->is(tok::semi)) {
2632       nextToken();
2633       addUnwrappedLine();
2634       break;
2635     } else if (FormatTok->is(tok::r_brace)) {
2636       addUnwrappedLine();
2637       break;
2638     } else {
2639       nextToken();
2640     }
2641   }
2642 
2643   // Parse the class body after the enum's ";" if any.
2644   parseLevel(/*HasOpeningBrace=*/true);
2645   nextToken();
2646   --Line->Level;
2647   addUnwrappedLine();
2648 }
2649 
parseRecord(bool ParseAsExpr)2650 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) {
2651   const FormatToken &InitialToken = *FormatTok;
2652   nextToken();
2653 
2654   // The actual identifier can be a nested name specifier, and in macros
2655   // it is often token-pasted.
2656   // An [[attribute]] can be before the identifier.
2657   while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash,
2658                             tok::kw___attribute, tok::kw___declspec,
2659                             tok::kw_alignas, tok::l_square, tok::r_square) ||
2660          ((Style.Language == FormatStyle::LK_Java ||
2661            Style.Language == FormatStyle::LK_JavaScript) &&
2662           FormatTok->isOneOf(tok::period, tok::comma))) {
2663     if (Style.Language == FormatStyle::LK_JavaScript &&
2664         FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) {
2665       // JavaScript/TypeScript supports inline object types in
2666       // extends/implements positions:
2667       //     class Foo implements {bar: number} { }
2668       nextToken();
2669       if (FormatTok->is(tok::l_brace)) {
2670         tryToParseBracedList();
2671         continue;
2672       }
2673     }
2674     bool IsNonMacroIdentifier =
2675         FormatTok->is(tok::identifier) &&
2676         FormatTok->TokenText != FormatTok->TokenText.upper();
2677     nextToken();
2678     // We can have macros or attributes in between 'class' and the class name.
2679     if (!IsNonMacroIdentifier) {
2680       if (FormatTok->Tok.is(tok::l_paren)) {
2681         parseParens();
2682       } else if (FormatTok->is(TT_AttributeSquare)) {
2683         parseSquare();
2684         // Consume the closing TT_AttributeSquare.
2685         if (FormatTok->Next && FormatTok->is(TT_AttributeSquare))
2686           nextToken();
2687       }
2688     }
2689   }
2690 
2691   // Note that parsing away template declarations here leads to incorrectly
2692   // accepting function declarations as record declarations.
2693   // In general, we cannot solve this problem. Consider:
2694   // class A<int> B() {}
2695   // which can be a function definition or a class definition when B() is a
2696   // macro. If we find enough real-world cases where this is a problem, we
2697   // can parse for the 'template' keyword in the beginning of the statement,
2698   // and thus rule out the record production in case there is no template
2699   // (this would still leave us with an ambiguity between template function
2700   // and class declarations).
2701   if (FormatTok->isOneOf(tok::colon, tok::less)) {
2702     while (!eof()) {
2703       if (FormatTok->is(tok::l_brace)) {
2704         calculateBraceTypes(/*ExpectClassBody=*/true);
2705         if (!tryToParseBracedList())
2706           break;
2707       }
2708       if (FormatTok->Tok.is(tok::semi))
2709         return;
2710       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) {
2711         addUnwrappedLine();
2712         nextToken();
2713         parseCSharpGenericTypeConstraint();
2714         break;
2715       }
2716       nextToken();
2717     }
2718   }
2719   if (FormatTok->Tok.is(tok::l_brace)) {
2720     if (ParseAsExpr) {
2721       parseChildBlock();
2722     } else {
2723       if (ShouldBreakBeforeBrace(Style, InitialToken))
2724         addUnwrappedLine();
2725 
2726       unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u;
2727       parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false);
2728     }
2729   }
2730   // There is no addUnwrappedLine() here so that we fall through to parsing a
2731   // structural element afterwards. Thus, in "class A {} n, m;",
2732   // "} n, m;" will end up in one unwrapped line.
2733 }
2734 
parseObjCMethod()2735 void UnwrappedLineParser::parseObjCMethod() {
2736   assert(FormatTok->Tok.isOneOf(tok::l_paren, tok::identifier) &&
2737          "'(' or identifier expected.");
2738   do {
2739     if (FormatTok->Tok.is(tok::semi)) {
2740       nextToken();
2741       addUnwrappedLine();
2742       return;
2743     } else if (FormatTok->Tok.is(tok::l_brace)) {
2744       if (Style.BraceWrapping.AfterFunction)
2745         addUnwrappedLine();
2746       parseBlock(/*MustBeDeclaration=*/false);
2747       addUnwrappedLine();
2748       return;
2749     } else {
2750       nextToken();
2751     }
2752   } while (!eof());
2753 }
2754 
parseObjCProtocolList()2755 void UnwrappedLineParser::parseObjCProtocolList() {
2756   assert(FormatTok->Tok.is(tok::less) && "'<' expected.");
2757   do {
2758     nextToken();
2759     // Early exit in case someone forgot a close angle.
2760     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
2761         FormatTok->Tok.isObjCAtKeyword(tok::objc_end))
2762       return;
2763   } while (!eof() && FormatTok->Tok.isNot(tok::greater));
2764   nextToken(); // Skip '>'.
2765 }
2766 
parseObjCUntilAtEnd()2767 void UnwrappedLineParser::parseObjCUntilAtEnd() {
2768   do {
2769     if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) {
2770       nextToken();
2771       addUnwrappedLine();
2772       break;
2773     }
2774     if (FormatTok->is(tok::l_brace)) {
2775       parseBlock(/*MustBeDeclaration=*/false);
2776       // In ObjC interfaces, nothing should be following the "}".
2777       addUnwrappedLine();
2778     } else if (FormatTok->is(tok::r_brace)) {
2779       // Ignore stray "}". parseStructuralElement doesn't consume them.
2780       nextToken();
2781       addUnwrappedLine();
2782     } else if (FormatTok->isOneOf(tok::minus, tok::plus)) {
2783       nextToken();
2784       parseObjCMethod();
2785     } else {
2786       parseStructuralElement();
2787     }
2788   } while (!eof());
2789 }
2790 
parseObjCInterfaceOrImplementation()2791 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() {
2792   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface ||
2793          FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation);
2794   nextToken();
2795   nextToken(); // interface name
2796 
2797   // @interface can be followed by a lightweight generic
2798   // specialization list, then either a base class or a category.
2799   if (FormatTok->Tok.is(tok::less)) {
2800     parseObjCLightweightGenerics();
2801   }
2802   if (FormatTok->Tok.is(tok::colon)) {
2803     nextToken();
2804     nextToken(); // base class name
2805     // The base class can also have lightweight generics applied to it.
2806     if (FormatTok->Tok.is(tok::less)) {
2807       parseObjCLightweightGenerics();
2808     }
2809   } else if (FormatTok->Tok.is(tok::l_paren))
2810     // Skip category, if present.
2811     parseParens();
2812 
2813   if (FormatTok->Tok.is(tok::less))
2814     parseObjCProtocolList();
2815 
2816   if (FormatTok->Tok.is(tok::l_brace)) {
2817     if (Style.BraceWrapping.AfterObjCDeclaration)
2818       addUnwrappedLine();
2819     parseBlock(/*MustBeDeclaration=*/true);
2820   }
2821 
2822   // With instance variables, this puts '}' on its own line.  Without instance
2823   // variables, this ends the @interface line.
2824   addUnwrappedLine();
2825 
2826   parseObjCUntilAtEnd();
2827 }
2828 
parseObjCLightweightGenerics()2829 void UnwrappedLineParser::parseObjCLightweightGenerics() {
2830   assert(FormatTok->Tok.is(tok::less));
2831   // Unlike protocol lists, generic parameterizations support
2832   // nested angles:
2833   //
2834   // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> :
2835   //     NSObject <NSCopying, NSSecureCoding>
2836   //
2837   // so we need to count how many open angles we have left.
2838   unsigned NumOpenAngles = 1;
2839   do {
2840     nextToken();
2841     // Early exit in case someone forgot a close angle.
2842     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
2843         FormatTok->Tok.isObjCAtKeyword(tok::objc_end))
2844       break;
2845     if (FormatTok->Tok.is(tok::less))
2846       ++NumOpenAngles;
2847     else if (FormatTok->Tok.is(tok::greater)) {
2848       assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative");
2849       --NumOpenAngles;
2850     }
2851   } while (!eof() && NumOpenAngles != 0);
2852   nextToken(); // Skip '>'.
2853 }
2854 
2855 // Returns true for the declaration/definition form of @protocol,
2856 // false for the expression form.
parseObjCProtocol()2857 bool UnwrappedLineParser::parseObjCProtocol() {
2858   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol);
2859   nextToken();
2860 
2861   if (FormatTok->is(tok::l_paren))
2862     // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);".
2863     return false;
2864 
2865   // The definition/declaration form,
2866   // @protocol Foo
2867   // - (int)someMethod;
2868   // @end
2869 
2870   nextToken(); // protocol name
2871 
2872   if (FormatTok->Tok.is(tok::less))
2873     parseObjCProtocolList();
2874 
2875   // Check for protocol declaration.
2876   if (FormatTok->Tok.is(tok::semi)) {
2877     nextToken();
2878     addUnwrappedLine();
2879     return true;
2880   }
2881 
2882   addUnwrappedLine();
2883   parseObjCUntilAtEnd();
2884   return true;
2885 }
2886 
parseJavaScriptEs6ImportExport()2887 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
2888   bool IsImport = FormatTok->is(Keywords.kw_import);
2889   assert(IsImport || FormatTok->is(tok::kw_export));
2890   nextToken();
2891 
2892   // Consume the "default" in "export default class/function".
2893   if (FormatTok->is(tok::kw_default))
2894     nextToken();
2895 
2896   // Consume "async function", "function" and "default function", so that these
2897   // get parsed as free-standing JS functions, i.e. do not require a trailing
2898   // semicolon.
2899   if (FormatTok->is(Keywords.kw_async))
2900     nextToken();
2901   if (FormatTok->is(Keywords.kw_function)) {
2902     nextToken();
2903     return;
2904   }
2905 
2906   // For imports, `export *`, `export {...}`, consume the rest of the line up
2907   // to the terminating `;`. For everything else, just return and continue
2908   // parsing the structural element, i.e. the declaration or expression for
2909   // `export default`.
2910   if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) &&
2911       !FormatTok->isStringLiteral())
2912     return;
2913 
2914   while (!eof()) {
2915     if (FormatTok->is(tok::semi))
2916       return;
2917     if (Line->Tokens.empty()) {
2918       // Common issue: Automatic Semicolon Insertion wrapped the line, so the
2919       // import statement should terminate.
2920       return;
2921     }
2922     if (FormatTok->is(tok::l_brace)) {
2923       FormatTok->setBlockKind(BK_Block);
2924       nextToken();
2925       parseBracedList();
2926     } else {
2927       nextToken();
2928     }
2929   }
2930 }
2931 
parseStatementMacro()2932 void UnwrappedLineParser::parseStatementMacro() {
2933   nextToken();
2934   if (FormatTok->is(tok::l_paren))
2935     parseParens();
2936   if (FormatTok->is(tok::semi))
2937     nextToken();
2938   addUnwrappedLine();
2939 }
2940 
printDebugInfo(const UnwrappedLine & Line,StringRef Prefix="")2941 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line,
2942                                                  StringRef Prefix = "") {
2943   llvm::dbgs() << Prefix << "Line(" << Line.Level
2944                << ", FSC=" << Line.FirstStartColumn << ")"
2945                << (Line.InPPDirective ? " MACRO" : "") << ": ";
2946   for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
2947                                                     E = Line.Tokens.end();
2948        I != E; ++I) {
2949     llvm::dbgs() << I->Tok->Tok.getName() << "["
2950                  << "T=" << (unsigned)I->Tok->getType()
2951                  << ", OC=" << I->Tok->OriginalColumn << "] ";
2952   }
2953   for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
2954                                                     E = Line.Tokens.end();
2955        I != E; ++I) {
2956     const UnwrappedLineNode &Node = *I;
2957     for (SmallVectorImpl<UnwrappedLine>::const_iterator
2958              I = Node.Children.begin(),
2959              E = Node.Children.end();
2960          I != E; ++I) {
2961       printDebugInfo(*I, "\nChild: ");
2962     }
2963   }
2964   llvm::dbgs() << "\n";
2965 }
2966 
addUnwrappedLine(LineLevel AdjustLevel)2967 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) {
2968   if (Line->Tokens.empty())
2969     return;
2970   LLVM_DEBUG({
2971     if (CurrentLines == &Lines)
2972       printDebugInfo(*Line);
2973   });
2974 
2975   // If this line closes a block when in Whitesmiths mode, remember that
2976   // information so that the level can be decreased after the line is added.
2977   // This has to happen after the addition of the line since the line itself
2978   // needs to be indented.
2979   bool ClosesWhitesmithsBlock =
2980       Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex &&
2981       Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
2982 
2983   CurrentLines->push_back(std::move(*Line));
2984   Line->Tokens.clear();
2985   Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex;
2986   Line->FirstStartColumn = 0;
2987 
2988   if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove)
2989     --Line->Level;
2990   if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) {
2991     CurrentLines->append(
2992         std::make_move_iterator(PreprocessorDirectives.begin()),
2993         std::make_move_iterator(PreprocessorDirectives.end()));
2994     PreprocessorDirectives.clear();
2995   }
2996   // Disconnect the current token from the last token on the previous line.
2997   FormatTok->Previous = nullptr;
2998 }
2999 
eof() const3000 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); }
3001 
isOnNewLine(const FormatToken & FormatTok)3002 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) {
3003   return (Line->InPPDirective || FormatTok.HasUnescapedNewline) &&
3004          FormatTok.NewlinesBefore > 0;
3005 }
3006 
3007 // Checks if \p FormatTok is a line comment that continues the line comment
3008 // section on \p Line.
3009 static bool
continuesLineCommentSection(const FormatToken & FormatTok,const UnwrappedLine & Line,const llvm::Regex & CommentPragmasRegex)3010 continuesLineCommentSection(const FormatToken &FormatTok,
3011                             const UnwrappedLine &Line,
3012                             const llvm::Regex &CommentPragmasRegex) {
3013   if (Line.Tokens.empty())
3014     return false;
3015 
3016   StringRef IndentContent = FormatTok.TokenText;
3017   if (FormatTok.TokenText.startswith("//") ||
3018       FormatTok.TokenText.startswith("/*"))
3019     IndentContent = FormatTok.TokenText.substr(2);
3020   if (CommentPragmasRegex.match(IndentContent))
3021     return false;
3022 
3023   // If Line starts with a line comment, then FormatTok continues the comment
3024   // section if its original column is greater or equal to the original start
3025   // column of the line.
3026   //
3027   // Define the min column token of a line as follows: if a line ends in '{' or
3028   // contains a '{' followed by a line comment, then the min column token is
3029   // that '{'. Otherwise, the min column token of the line is the first token of
3030   // the line.
3031   //
3032   // If Line starts with a token other than a line comment, then FormatTok
3033   // continues the comment section if its original column is greater than the
3034   // original start column of the min column token of the line.
3035   //
3036   // For example, the second line comment continues the first in these cases:
3037   //
3038   // // first line
3039   // // second line
3040   //
3041   // and:
3042   //
3043   // // first line
3044   //  // second line
3045   //
3046   // and:
3047   //
3048   // int i; // first line
3049   //  // second line
3050   //
3051   // and:
3052   //
3053   // do { // first line
3054   //      // second line
3055   //   int i;
3056   // } while (true);
3057   //
3058   // and:
3059   //
3060   // enum {
3061   //   a, // first line
3062   //    // second line
3063   //   b
3064   // };
3065   //
3066   // The second line comment doesn't continue the first in these cases:
3067   //
3068   //   // first line
3069   //  // second line
3070   //
3071   // and:
3072   //
3073   // int i; // first line
3074   // // second line
3075   //
3076   // and:
3077   //
3078   // do { // first line
3079   //   // second line
3080   //   int i;
3081   // } while (true);
3082   //
3083   // and:
3084   //
3085   // enum {
3086   //   a, // first line
3087   //   // second line
3088   // };
3089   const FormatToken *MinColumnToken = Line.Tokens.front().Tok;
3090 
3091   // Scan for '{//'. If found, use the column of '{' as a min column for line
3092   // comment section continuation.
3093   const FormatToken *PreviousToken = nullptr;
3094   for (const UnwrappedLineNode &Node : Line.Tokens) {
3095     if (PreviousToken && PreviousToken->is(tok::l_brace) &&
3096         isLineComment(*Node.Tok)) {
3097       MinColumnToken = PreviousToken;
3098       break;
3099     }
3100     PreviousToken = Node.Tok;
3101 
3102     // Grab the last newline preceding a token in this unwrapped line.
3103     if (Node.Tok->NewlinesBefore > 0) {
3104       MinColumnToken = Node.Tok;
3105     }
3106   }
3107   if (PreviousToken && PreviousToken->is(tok::l_brace)) {
3108     MinColumnToken = PreviousToken;
3109   }
3110 
3111   return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok,
3112                               MinColumnToken);
3113 }
3114 
flushComments(bool NewlineBeforeNext)3115 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
3116   bool JustComments = Line->Tokens.empty();
3117   for (SmallVectorImpl<FormatToken *>::const_iterator
3118            I = CommentsBeforeNextToken.begin(),
3119            E = CommentsBeforeNextToken.end();
3120        I != E; ++I) {
3121     // Line comments that belong to the same line comment section are put on the
3122     // same line since later we might want to reflow content between them.
3123     // Additional fine-grained breaking of line comment sections is controlled
3124     // by the class BreakableLineCommentSection in case it is desirable to keep
3125     // several line comment sections in the same unwrapped line.
3126     //
3127     // FIXME: Consider putting separate line comment sections as children to the
3128     // unwrapped line instead.
3129     (*I)->ContinuesLineCommentSection =
3130         continuesLineCommentSection(**I, *Line, CommentPragmasRegex);
3131     if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection)
3132       addUnwrappedLine();
3133     pushToken(*I);
3134   }
3135   if (NewlineBeforeNext && JustComments)
3136     addUnwrappedLine();
3137   CommentsBeforeNextToken.clear();
3138 }
3139 
nextToken(int LevelDifference)3140 void UnwrappedLineParser::nextToken(int LevelDifference) {
3141   if (eof())
3142     return;
3143   flushComments(isOnNewLine(*FormatTok));
3144   pushToken(FormatTok);
3145   FormatToken *Previous = FormatTok;
3146   if (Style.Language != FormatStyle::LK_JavaScript)
3147     readToken(LevelDifference);
3148   else
3149     readTokenWithJavaScriptASI();
3150   FormatTok->Previous = Previous;
3151 }
3152 
distributeComments(const SmallVectorImpl<FormatToken * > & Comments,const FormatToken * NextTok)3153 void UnwrappedLineParser::distributeComments(
3154     const SmallVectorImpl<FormatToken *> &Comments,
3155     const FormatToken *NextTok) {
3156   // Whether or not a line comment token continues a line is controlled by
3157   // the method continuesLineCommentSection, with the following caveat:
3158   //
3159   // Define a trail of Comments to be a nonempty proper postfix of Comments such
3160   // that each comment line from the trail is aligned with the next token, if
3161   // the next token exists. If a trail exists, the beginning of the maximal
3162   // trail is marked as a start of a new comment section.
3163   //
3164   // For example in this code:
3165   //
3166   // int a; // line about a
3167   //   // line 1 about b
3168   //   // line 2 about b
3169   //   int b;
3170   //
3171   // the two lines about b form a maximal trail, so there are two sections, the
3172   // first one consisting of the single comment "// line about a" and the
3173   // second one consisting of the next two comments.
3174   if (Comments.empty())
3175     return;
3176   bool ShouldPushCommentsInCurrentLine = true;
3177   bool HasTrailAlignedWithNextToken = false;
3178   unsigned StartOfTrailAlignedWithNextToken = 0;
3179   if (NextTok) {
3180     // We are skipping the first element intentionally.
3181     for (unsigned i = Comments.size() - 1; i > 0; --i) {
3182       if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) {
3183         HasTrailAlignedWithNextToken = true;
3184         StartOfTrailAlignedWithNextToken = i;
3185       }
3186     }
3187   }
3188   for (unsigned i = 0, e = Comments.size(); i < e; ++i) {
3189     FormatToken *FormatTok = Comments[i];
3190     if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) {
3191       FormatTok->ContinuesLineCommentSection = false;
3192     } else {
3193       FormatTok->ContinuesLineCommentSection =
3194           continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex);
3195     }
3196     if (!FormatTok->ContinuesLineCommentSection &&
3197         (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) {
3198       ShouldPushCommentsInCurrentLine = false;
3199     }
3200     if (ShouldPushCommentsInCurrentLine) {
3201       pushToken(FormatTok);
3202     } else {
3203       CommentsBeforeNextToken.push_back(FormatTok);
3204     }
3205   }
3206 }
3207 
readToken(int LevelDifference)3208 void UnwrappedLineParser::readToken(int LevelDifference) {
3209   SmallVector<FormatToken *, 1> Comments;
3210   do {
3211     FormatTok = Tokens->getNextToken();
3212     assert(FormatTok);
3213     while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) &&
3214            (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) {
3215       distributeComments(Comments, FormatTok);
3216       Comments.clear();
3217       // If there is an unfinished unwrapped line, we flush the preprocessor
3218       // directives only after that unwrapped line was finished later.
3219       bool SwitchToPreprocessorLines = !Line->Tokens.empty();
3220       ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
3221       assert((LevelDifference >= 0 ||
3222               static_cast<unsigned>(-LevelDifference) <= Line->Level) &&
3223              "LevelDifference makes Line->Level negative");
3224       Line->Level += LevelDifference;
3225       // Comments stored before the preprocessor directive need to be output
3226       // before the preprocessor directive, at the same level as the
3227       // preprocessor directive, as we consider them to apply to the directive.
3228       if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash &&
3229           PPBranchLevel > 0)
3230         Line->Level += PPBranchLevel;
3231       flushComments(isOnNewLine(*FormatTok));
3232       parsePPDirective();
3233     }
3234     while (FormatTok->getType() == TT_ConflictStart ||
3235            FormatTok->getType() == TT_ConflictEnd ||
3236            FormatTok->getType() == TT_ConflictAlternative) {
3237       if (FormatTok->getType() == TT_ConflictStart) {
3238         conditionalCompilationStart(/*Unreachable=*/false);
3239       } else if (FormatTok->getType() == TT_ConflictAlternative) {
3240         conditionalCompilationAlternative();
3241       } else if (FormatTok->getType() == TT_ConflictEnd) {
3242         conditionalCompilationEnd();
3243       }
3244       FormatTok = Tokens->getNextToken();
3245       FormatTok->MustBreakBefore = true;
3246     }
3247 
3248     if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) &&
3249         !Line->InPPDirective) {
3250       continue;
3251     }
3252 
3253     if (!FormatTok->Tok.is(tok::comment)) {
3254       distributeComments(Comments, FormatTok);
3255       Comments.clear();
3256       return;
3257     }
3258 
3259     Comments.push_back(FormatTok);
3260   } while (!eof());
3261 
3262   distributeComments(Comments, nullptr);
3263   Comments.clear();
3264 }
3265 
pushToken(FormatToken * Tok)3266 void UnwrappedLineParser::pushToken(FormatToken *Tok) {
3267   Line->Tokens.push_back(UnwrappedLineNode(Tok));
3268   if (MustBreakBeforeNextToken) {
3269     Line->Tokens.back().Tok->MustBreakBefore = true;
3270     MustBreakBeforeNextToken = false;
3271   }
3272 }
3273 
3274 } // end namespace format
3275 } // end namespace clang
3276