1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 /// 10 /// \file 11 /// \brief This file contains the implementation of the UnwrappedLineParser, 12 /// which turns a stream of tokens into UnwrappedLines. 13 /// 14 //===----------------------------------------------------------------------===// 15 16 #include "UnwrappedLineParser.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/Support/Debug.h" 19 #include "llvm/Support/raw_ostream.h" 20 21 #define DEBUG_TYPE "format-parser" 22 23 namespace clang { 24 namespace format { 25 26 class FormatTokenSource { 27 public: 28 virtual ~FormatTokenSource() {} 29 virtual FormatToken *getNextToken() = 0; 30 31 virtual unsigned getPosition() = 0; 32 virtual FormatToken *setPosition(unsigned Position) = 0; 33 }; 34 35 namespace { 36 37 class ScopedDeclarationState { 38 public: 39 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 40 bool MustBeDeclaration) 41 : Line(Line), Stack(Stack) { 42 Line.MustBeDeclaration = MustBeDeclaration; 43 Stack.push_back(MustBeDeclaration); 44 } 45 ~ScopedDeclarationState() { 46 Stack.pop_back(); 47 if (!Stack.empty()) 48 Line.MustBeDeclaration = Stack.back(); 49 else 50 Line.MustBeDeclaration = true; 51 } 52 53 private: 54 UnwrappedLine &Line; 55 std::vector<bool> &Stack; 56 }; 57 58 static bool isLineComment(const FormatToken &FormatTok) { 59 return FormatTok.is(tok::comment) && 60 FormatTok.TokenText.startswith("//"); 61 } 62 63 // Checks if \p FormatTok is a line comment that continues the line comment 64 // \p Previous. The original column of \p MinColumnToken is used to determine 65 // whether \p FormatTok is indented enough to the right to continue \p Previous. 66 static bool continuesLineComment(const FormatToken &FormatTok, 67 const FormatToken *Previous, 68 const FormatToken *MinColumnToken) { 69 if (!Previous || !MinColumnToken) 70 return false; 71 unsigned MinContinueColumn = 72 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 73 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 74 isLineComment(*Previous) && 75 FormatTok.OriginalColumn >= MinContinueColumn; 76 } 77 78 class ScopedMacroState : public FormatTokenSource { 79 public: 80 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 81 FormatToken *&ResetToken) 82 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 83 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 84 Token(nullptr), PreviousToken(nullptr) { 85 TokenSource = this; 86 Line.Level = 0; 87 Line.InPPDirective = true; 88 } 89 90 ~ScopedMacroState() override { 91 TokenSource = PreviousTokenSource; 92 ResetToken = Token; 93 Line.InPPDirective = false; 94 Line.Level = PreviousLineLevel; 95 } 96 97 FormatToken *getNextToken() override { 98 // The \c UnwrappedLineParser guards against this by never calling 99 // \c getNextToken() after it has encountered the first eof token. 100 assert(!eof()); 101 PreviousToken = Token; 102 Token = PreviousTokenSource->getNextToken(); 103 if (eof()) 104 return getFakeEOF(); 105 return Token; 106 } 107 108 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 109 110 FormatToken *setPosition(unsigned Position) override { 111 PreviousToken = nullptr; 112 Token = PreviousTokenSource->setPosition(Position); 113 return Token; 114 } 115 116 private: 117 bool eof() { 118 return Token && Token->HasUnescapedNewline && 119 !continuesLineComment(*Token, PreviousToken, 120 /*MinColumnToken=*/PreviousToken); 121 } 122 123 FormatToken *getFakeEOF() { 124 static bool EOFInitialized = false; 125 static FormatToken FormatTok; 126 if (!EOFInitialized) { 127 FormatTok.Tok.startToken(); 128 FormatTok.Tok.setKind(tok::eof); 129 EOFInitialized = true; 130 } 131 return &FormatTok; 132 } 133 134 UnwrappedLine &Line; 135 FormatTokenSource *&TokenSource; 136 FormatToken *&ResetToken; 137 unsigned PreviousLineLevel; 138 FormatTokenSource *PreviousTokenSource; 139 140 FormatToken *Token; 141 FormatToken *PreviousToken; 142 }; 143 144 } // end anonymous namespace 145 146 class ScopedLineState { 147 public: 148 ScopedLineState(UnwrappedLineParser &Parser, 149 bool SwitchToPreprocessorLines = false) 150 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 151 if (SwitchToPreprocessorLines) 152 Parser.CurrentLines = &Parser.PreprocessorDirectives; 153 else if (!Parser.Line->Tokens.empty()) 154 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 155 PreBlockLine = std::move(Parser.Line); 156 Parser.Line = llvm::make_unique<UnwrappedLine>(); 157 Parser.Line->Level = PreBlockLine->Level; 158 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 159 } 160 161 ~ScopedLineState() { 162 if (!Parser.Line->Tokens.empty()) { 163 Parser.addUnwrappedLine(); 164 } 165 assert(Parser.Line->Tokens.empty()); 166 Parser.Line = std::move(PreBlockLine); 167 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 168 Parser.MustBreakBeforeNextToken = true; 169 Parser.CurrentLines = OriginalLines; 170 } 171 172 private: 173 UnwrappedLineParser &Parser; 174 175 std::unique_ptr<UnwrappedLine> PreBlockLine; 176 SmallVectorImpl<UnwrappedLine> *OriginalLines; 177 }; 178 179 class CompoundStatementIndenter { 180 public: 181 CompoundStatementIndenter(UnwrappedLineParser *Parser, 182 const FormatStyle &Style, unsigned &LineLevel) 183 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 184 if (Style.BraceWrapping.AfterControlStatement) 185 Parser->addUnwrappedLine(); 186 if (Style.BraceWrapping.IndentBraces) 187 ++LineLevel; 188 } 189 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 190 191 private: 192 unsigned &LineLevel; 193 unsigned OldLineLevel; 194 }; 195 196 namespace { 197 198 class IndexedTokenSource : public FormatTokenSource { 199 public: 200 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 201 : Tokens(Tokens), Position(-1) {} 202 203 FormatToken *getNextToken() override { 204 ++Position; 205 return Tokens[Position]; 206 } 207 208 unsigned getPosition() override { 209 assert(Position >= 0); 210 return Position; 211 } 212 213 FormatToken *setPosition(unsigned P) override { 214 Position = P; 215 return Tokens[Position]; 216 } 217 218 void reset() { Position = -1; } 219 220 private: 221 ArrayRef<FormatToken *> Tokens; 222 int Position; 223 }; 224 225 } // end anonymous namespace 226 227 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 228 const AdditionalKeywords &Keywords, 229 ArrayRef<FormatToken *> Tokens, 230 UnwrappedLineConsumer &Callback) 231 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 232 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 233 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 234 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {} 235 236 void UnwrappedLineParser::reset() { 237 PPBranchLevel = -1; 238 Line.reset(new UnwrappedLine); 239 CommentsBeforeNextToken.clear(); 240 FormatTok = nullptr; 241 MustBreakBeforeNextToken = false; 242 PreprocessorDirectives.clear(); 243 CurrentLines = &Lines; 244 DeclarationScopeStack.clear(); 245 PPStack.clear(); 246 } 247 248 void UnwrappedLineParser::parse() { 249 IndexedTokenSource TokenSource(AllTokens); 250 do { 251 DEBUG(llvm::dbgs() << "----\n"); 252 reset(); 253 Tokens = &TokenSource; 254 TokenSource.reset(); 255 256 readToken(); 257 parseFile(); 258 // Create line with eof token. 259 pushToken(FormatTok); 260 addUnwrappedLine(); 261 262 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 263 E = Lines.end(); 264 I != E; ++I) { 265 Callback.consumeUnwrappedLine(*I); 266 } 267 Callback.finishRun(); 268 Lines.clear(); 269 while (!PPLevelBranchIndex.empty() && 270 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 271 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 272 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 273 } 274 if (!PPLevelBranchIndex.empty()) { 275 ++PPLevelBranchIndex.back(); 276 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 277 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 278 } 279 } while (!PPLevelBranchIndex.empty()); 280 } 281 282 void UnwrappedLineParser::parseFile() { 283 // The top-level context in a file always has declarations, except for pre- 284 // processor directives and JavaScript files. 285 bool MustBeDeclaration = 286 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 287 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 288 MustBeDeclaration); 289 if (Style.Language == FormatStyle::LK_TextProto) 290 parseBracedList(); 291 else 292 parseLevel(/*HasOpeningBrace=*/false); 293 // Make sure to format the remaining tokens. 294 flushComments(true); 295 addUnwrappedLine(); 296 } 297 298 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 299 bool SwitchLabelEncountered = false; 300 do { 301 tok::TokenKind kind = FormatTok->Tok.getKind(); 302 if (FormatTok->Type == TT_MacroBlockBegin) { 303 kind = tok::l_brace; 304 } else if (FormatTok->Type == TT_MacroBlockEnd) { 305 kind = tok::r_brace; 306 } 307 308 switch (kind) { 309 case tok::comment: 310 nextToken(); 311 addUnwrappedLine(); 312 break; 313 case tok::l_brace: 314 // FIXME: Add parameter whether this can happen - if this happens, we must 315 // be in a non-declaration context. 316 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 317 continue; 318 parseBlock(/*MustBeDeclaration=*/false); 319 addUnwrappedLine(); 320 break; 321 case tok::r_brace: 322 if (HasOpeningBrace) 323 return; 324 nextToken(); 325 addUnwrappedLine(); 326 break; 327 case tok::kw_default: 328 case tok::kw_case: 329 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) { 330 // A 'case: string' style field declaration. 331 parseStructuralElement(); 332 break; 333 } 334 if (!SwitchLabelEncountered && 335 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 336 ++Line->Level; 337 SwitchLabelEncountered = true; 338 parseStructuralElement(); 339 break; 340 default: 341 parseStructuralElement(); 342 break; 343 } 344 } while (!eof()); 345 } 346 347 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 348 // We'll parse forward through the tokens until we hit 349 // a closing brace or eof - note that getNextToken() will 350 // parse macros, so this will magically work inside macro 351 // definitions, too. 352 unsigned StoredPosition = Tokens->getPosition(); 353 FormatToken *Tok = FormatTok; 354 const FormatToken *PrevTok = getPreviousToken(); 355 // Keep a stack of positions of lbrace tokens. We will 356 // update information about whether an lbrace starts a 357 // braced init list or a different block during the loop. 358 SmallVector<FormatToken *, 8> LBraceStack; 359 assert(Tok->Tok.is(tok::l_brace)); 360 do { 361 // Get next non-comment token. 362 FormatToken *NextTok; 363 unsigned ReadTokens = 0; 364 do { 365 NextTok = Tokens->getNextToken(); 366 ++ReadTokens; 367 } while (NextTok->is(tok::comment)); 368 369 switch (Tok->Tok.getKind()) { 370 case tok::l_brace: 371 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok) { 372 if (PrevTok->is(tok::colon)) 373 // A colon indicates this code is in a type, or a braced list 374 // following a label in an object literal ({a: {b: 1}}). The code 375 // below could be confused by semicolons between the individual 376 // members in a type member list, which would normally trigger 377 // BK_Block. In both cases, this must be parsed as an inline braced 378 // init. 379 Tok->BlockKind = BK_BracedInit; 380 else if (PrevTok->is(tok::r_paren)) 381 // `) { }` can only occur in function or method declarations in JS. 382 Tok->BlockKind = BK_Block; 383 } else { 384 Tok->BlockKind = BK_Unknown; 385 } 386 LBraceStack.push_back(Tok); 387 break; 388 case tok::r_brace: 389 if (LBraceStack.empty()) 390 break; 391 if (LBraceStack.back()->BlockKind == BK_Unknown) { 392 bool ProbablyBracedList = false; 393 if (Style.Language == FormatStyle::LK_Proto) { 394 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 395 } else { 396 // Using OriginalColumn to distinguish between ObjC methods and 397 // binary operators is a bit hacky. 398 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 399 NextTok->OriginalColumn == 0; 400 401 // If there is a comma, semicolon or right paren after the closing 402 // brace, we assume this is a braced initializer list. Note that 403 // regardless how we mark inner braces here, we will overwrite the 404 // BlockKind later if we parse a braced list (where all blocks 405 // inside are by default braced lists), or when we explicitly detect 406 // blocks (for example while parsing lambdas). 407 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 408 // braced list in JS. 409 ProbablyBracedList = 410 (Style.Language == FormatStyle::LK_JavaScript && 411 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 412 Keywords.kw_as)) || 413 (Style.isCpp() && NextTok->is(tok::l_paren)) || 414 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 415 tok::r_paren, tok::r_square, tok::l_brace, 416 tok::l_square, tok::ellipsis) || 417 (NextTok->is(tok::identifier) && 418 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 419 (NextTok->is(tok::semi) && 420 (!ExpectClassBody || LBraceStack.size() != 1)) || 421 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 422 } 423 if (ProbablyBracedList) { 424 Tok->BlockKind = BK_BracedInit; 425 LBraceStack.back()->BlockKind = BK_BracedInit; 426 } else { 427 Tok->BlockKind = BK_Block; 428 LBraceStack.back()->BlockKind = BK_Block; 429 } 430 } 431 LBraceStack.pop_back(); 432 break; 433 case tok::at: 434 case tok::semi: 435 case tok::kw_if: 436 case tok::kw_while: 437 case tok::kw_for: 438 case tok::kw_switch: 439 case tok::kw_try: 440 case tok::kw___try: 441 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 442 LBraceStack.back()->BlockKind = BK_Block; 443 break; 444 default: 445 break; 446 } 447 PrevTok = Tok; 448 Tok = NextTok; 449 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 450 451 // Assume other blocks for all unclosed opening braces. 452 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 453 if (LBraceStack[i]->BlockKind == BK_Unknown) 454 LBraceStack[i]->BlockKind = BK_Block; 455 } 456 457 FormatTok = Tokens->setPosition(StoredPosition); 458 } 459 460 template <class T> 461 static inline void hash_combine(std::size_t &seed, const T &v) { 462 std::hash<T> hasher; 463 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 464 } 465 466 size_t UnwrappedLineParser::computePPHash() const { 467 size_t h = 0; 468 for (const auto &i : PPStack) { 469 hash_combine(h, size_t(i.Kind)); 470 hash_combine(h, i.Line); 471 } 472 return h; 473 } 474 475 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 476 bool MunchSemi) { 477 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 478 "'{' or macro block token expected"); 479 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 480 FormatTok->BlockKind = BK_Block; 481 482 size_t PPStartHash = computePPHash(); 483 484 unsigned InitialLevel = Line->Level; 485 nextToken(/*LevelDifference=*/AddLevel ? 1 : 0); 486 487 if (MacroBlock && FormatTok->is(tok::l_paren)) 488 parseParens(); 489 490 size_t NbPreprocessorDirectives = 491 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 492 addUnwrappedLine(); 493 size_t OpeningLineIndex = 494 CurrentLines->empty() 495 ? (UnwrappedLine::kInvalidIndex) 496 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 497 498 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 499 MustBeDeclaration); 500 if (AddLevel) 501 ++Line->Level; 502 parseLevel(/*HasOpeningBrace=*/true); 503 504 if (eof()) 505 return; 506 507 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 508 : !FormatTok->is(tok::r_brace)) { 509 Line->Level = InitialLevel; 510 FormatTok->BlockKind = BK_Block; 511 return; 512 } 513 514 size_t PPEndHash = computePPHash(); 515 516 // Munch the closing brace. 517 nextToken(/*LevelDifference=*/AddLevel ? -1 : 0); 518 519 if (MacroBlock && FormatTok->is(tok::l_paren)) 520 parseParens(); 521 522 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 523 nextToken(); 524 Line->Level = InitialLevel; 525 526 if (PPStartHash == PPEndHash) { 527 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 528 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 529 // Update the opening line to add the forward reference as well 530 (*CurrentLines)[OpeningLineIndex].MatchingOpeningBlockLineIndex = 531 CurrentLines->size() - 1; 532 } 533 } 534 } 535 536 static bool isGoogScope(const UnwrappedLine &Line) { 537 // FIXME: Closure-library specific stuff should not be hard-coded but be 538 // configurable. 539 if (Line.Tokens.size() < 4) 540 return false; 541 auto I = Line.Tokens.begin(); 542 if (I->Tok->TokenText != "goog") 543 return false; 544 ++I; 545 if (I->Tok->isNot(tok::period)) 546 return false; 547 ++I; 548 if (I->Tok->TokenText != "scope") 549 return false; 550 ++I; 551 return I->Tok->is(tok::l_paren); 552 } 553 554 static bool isIIFE(const UnwrappedLine &Line, 555 const AdditionalKeywords &Keywords) { 556 // Look for the start of an immediately invoked anonymous function. 557 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 558 // This is commonly done in JavaScript to create a new, anonymous scope. 559 // Example: (function() { ... })() 560 if (Line.Tokens.size() < 3) 561 return false; 562 auto I = Line.Tokens.begin(); 563 if (I->Tok->isNot(tok::l_paren)) 564 return false; 565 ++I; 566 if (I->Tok->isNot(Keywords.kw_function)) 567 return false; 568 ++I; 569 return I->Tok->is(tok::l_paren); 570 } 571 572 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 573 const FormatToken &InitialToken) { 574 if (InitialToken.is(tok::kw_namespace)) 575 return Style.BraceWrapping.AfterNamespace; 576 if (InitialToken.is(tok::kw_class)) 577 return Style.BraceWrapping.AfterClass; 578 if (InitialToken.is(tok::kw_union)) 579 return Style.BraceWrapping.AfterUnion; 580 if (InitialToken.is(tok::kw_struct)) 581 return Style.BraceWrapping.AfterStruct; 582 return false; 583 } 584 585 void UnwrappedLineParser::parseChildBlock() { 586 FormatTok->BlockKind = BK_Block; 587 nextToken(); 588 { 589 bool SkipIndent = 590 (Style.Language == FormatStyle::LK_JavaScript && 591 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 592 ScopedLineState LineState(*this); 593 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 594 /*MustBeDeclaration=*/false); 595 Line->Level += SkipIndent ? 0 : 1; 596 parseLevel(/*HasOpeningBrace=*/true); 597 flushComments(isOnNewLine(*FormatTok)); 598 Line->Level -= SkipIndent ? 0 : 1; 599 } 600 nextToken(); 601 } 602 603 void UnwrappedLineParser::parsePPDirective() { 604 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 605 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 606 nextToken(); 607 608 if (!FormatTok->Tok.getIdentifierInfo()) { 609 parsePPUnknown(); 610 return; 611 } 612 613 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 614 case tok::pp_define: 615 parsePPDefine(); 616 return; 617 case tok::pp_if: 618 parsePPIf(/*IfDef=*/false); 619 break; 620 case tok::pp_ifdef: 621 case tok::pp_ifndef: 622 parsePPIf(/*IfDef=*/true); 623 break; 624 case tok::pp_else: 625 parsePPElse(); 626 break; 627 case tok::pp_elif: 628 parsePPElIf(); 629 break; 630 case tok::pp_endif: 631 parsePPEndIf(); 632 break; 633 default: 634 parsePPUnknown(); 635 break; 636 } 637 } 638 639 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 640 size_t Line = CurrentLines->size(); 641 if (CurrentLines == &PreprocessorDirectives) 642 Line += Lines.size(); 643 644 if (Unreachable || 645 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 646 PPStack.push_back({PP_Unreachable, Line}); 647 else 648 PPStack.push_back({PP_Conditional, Line}); 649 } 650 651 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 652 ++PPBranchLevel; 653 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 654 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 655 PPLevelBranchIndex.push_back(0); 656 PPLevelBranchCount.push_back(0); 657 } 658 PPChainBranchIndex.push(0); 659 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 660 conditionalCompilationCondition(Unreachable || Skip); 661 } 662 663 void UnwrappedLineParser::conditionalCompilationAlternative() { 664 if (!PPStack.empty()) 665 PPStack.pop_back(); 666 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 667 if (!PPChainBranchIndex.empty()) 668 ++PPChainBranchIndex.top(); 669 conditionalCompilationCondition( 670 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 671 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 672 } 673 674 void UnwrappedLineParser::conditionalCompilationEnd() { 675 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 676 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 677 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 678 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 679 } 680 } 681 // Guard against #endif's without #if. 682 if (PPBranchLevel > 0) 683 --PPBranchLevel; 684 if (!PPChainBranchIndex.empty()) 685 PPChainBranchIndex.pop(); 686 if (!PPStack.empty()) 687 PPStack.pop_back(); 688 } 689 690 void UnwrappedLineParser::parsePPIf(bool IfDef) { 691 bool IfNDef = FormatTok->is(tok::pp_ifndef); 692 nextToken(); 693 bool Unreachable = false; 694 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 695 Unreachable = true; 696 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 697 Unreachable = true; 698 conditionalCompilationStart(Unreachable); 699 parsePPUnknown(); 700 } 701 702 void UnwrappedLineParser::parsePPElse() { 703 conditionalCompilationAlternative(); 704 parsePPUnknown(); 705 } 706 707 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 708 709 void UnwrappedLineParser::parsePPEndIf() { 710 conditionalCompilationEnd(); 711 parsePPUnknown(); 712 } 713 714 void UnwrappedLineParser::parsePPDefine() { 715 nextToken(); 716 717 if (FormatTok->Tok.getKind() != tok::identifier) { 718 parsePPUnknown(); 719 return; 720 } 721 nextToken(); 722 if (FormatTok->Tok.getKind() == tok::l_paren && 723 FormatTok->WhitespaceRange.getBegin() == 724 FormatTok->WhitespaceRange.getEnd()) { 725 parseParens(); 726 } 727 addUnwrappedLine(); 728 Line->Level = 1; 729 730 // Errors during a preprocessor directive can only affect the layout of the 731 // preprocessor directive, and thus we ignore them. An alternative approach 732 // would be to use the same approach we use on the file level (no 733 // re-indentation if there was a structural error) within the macro 734 // definition. 735 parseFile(); 736 } 737 738 void UnwrappedLineParser::parsePPUnknown() { 739 do { 740 nextToken(); 741 } while (!eof()); 742 addUnwrappedLine(); 743 } 744 745 // Here we blacklist certain tokens that are not usually the first token in an 746 // unwrapped line. This is used in attempt to distinguish macro calls without 747 // trailing semicolons from other constructs split to several lines. 748 static bool tokenCanStartNewLine(const clang::Token &Tok) { 749 // Semicolon can be a null-statement, l_square can be a start of a macro or 750 // a C++11 attribute, but this doesn't seem to be common. 751 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 752 Tok.isNot(tok::l_square) && 753 // Tokens that can only be used as binary operators and a part of 754 // overloaded operator names. 755 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 756 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 757 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 758 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 759 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 760 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 761 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 762 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 763 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 764 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 765 Tok.isNot(tok::lesslessequal) && 766 // Colon is used in labels, base class lists, initializer lists, 767 // range-based for loops, ternary operator, but should never be the 768 // first token in an unwrapped line. 769 Tok.isNot(tok::colon) && 770 // 'noexcept' is a trailing annotation. 771 Tok.isNot(tok::kw_noexcept); 772 } 773 774 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 775 const FormatToken *FormatTok) { 776 // FIXME: This returns true for C/C++ keywords like 'struct'. 777 return FormatTok->is(tok::identifier) && 778 (FormatTok->Tok.getIdentifierInfo() == nullptr || 779 !FormatTok->isOneOf( 780 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 781 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 782 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 783 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 784 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 785 Keywords.kw_instanceof, Keywords.kw_interface, 786 Keywords.kw_throws, Keywords.kw_from)); 787 } 788 789 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 790 const FormatToken *FormatTok) { 791 return FormatTok->Tok.isLiteral() || 792 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 793 mustBeJSIdent(Keywords, FormatTok); 794 } 795 796 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 797 // when encountered after a value (see mustBeJSIdentOrValue). 798 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 799 const FormatToken *FormatTok) { 800 return FormatTok->isOneOf( 801 tok::kw_return, Keywords.kw_yield, 802 // conditionals 803 tok::kw_if, tok::kw_else, 804 // loops 805 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 806 // switch/case 807 tok::kw_switch, tok::kw_case, 808 // exceptions 809 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 810 // declaration 811 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 812 Keywords.kw_async, Keywords.kw_function, 813 // import/export 814 Keywords.kw_import, tok::kw_export); 815 } 816 817 // readTokenWithJavaScriptASI reads the next token and terminates the current 818 // line if JavaScript Automatic Semicolon Insertion must 819 // happen between the current token and the next token. 820 // 821 // This method is conservative - it cannot cover all edge cases of JavaScript, 822 // but only aims to correctly handle certain well known cases. It *must not* 823 // return true in speculative cases. 824 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 825 FormatToken *Previous = FormatTok; 826 readToken(); 827 FormatToken *Next = FormatTok; 828 829 bool IsOnSameLine = 830 CommentsBeforeNextToken.empty() 831 ? Next->NewlinesBefore == 0 832 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 833 if (IsOnSameLine) 834 return; 835 836 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 837 bool PreviousStartsTemplateExpr = 838 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 839 if (PreviousMustBeValue && Line && Line->Tokens.size() > 1) { 840 // If the token before the previous one is an '@', the previous token is an 841 // annotation and can precede another identifier/value. 842 const FormatToken *PrePrevious = std::prev(Line->Tokens.end(), 2)->Tok; 843 if (PrePrevious->is(tok::at)) 844 return; 845 } 846 if (Next->is(tok::exclaim) && PreviousMustBeValue) 847 return addUnwrappedLine(); 848 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 849 bool NextEndsTemplateExpr = 850 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 851 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 852 (PreviousMustBeValue || 853 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 854 tok::minusminus))) 855 return addUnwrappedLine(); 856 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 857 isJSDeclOrStmt(Keywords, Next)) 858 return addUnwrappedLine(); 859 } 860 861 void UnwrappedLineParser::parseStructuralElement() { 862 assert(!FormatTok->is(tok::l_brace)); 863 if (Style.Language == FormatStyle::LK_TableGen && 864 FormatTok->is(tok::pp_include)) { 865 nextToken(); 866 if (FormatTok->is(tok::string_literal)) 867 nextToken(); 868 addUnwrappedLine(); 869 return; 870 } 871 switch (FormatTok->Tok.getKind()) { 872 case tok::at: 873 nextToken(); 874 if (FormatTok->Tok.is(tok::l_brace)) { 875 nextToken(); 876 parseBracedList(); 877 break; 878 } 879 switch (FormatTok->Tok.getObjCKeywordID()) { 880 case tok::objc_public: 881 case tok::objc_protected: 882 case tok::objc_package: 883 case tok::objc_private: 884 return parseAccessSpecifier(); 885 case tok::objc_interface: 886 case tok::objc_implementation: 887 return parseObjCInterfaceOrImplementation(); 888 case tok::objc_protocol: 889 return parseObjCProtocol(); 890 case tok::objc_end: 891 return; // Handled by the caller. 892 case tok::objc_optional: 893 case tok::objc_required: 894 nextToken(); 895 addUnwrappedLine(); 896 return; 897 case tok::objc_autoreleasepool: 898 nextToken(); 899 if (FormatTok->Tok.is(tok::l_brace)) { 900 if (Style.BraceWrapping.AfterObjCDeclaration) 901 addUnwrappedLine(); 902 parseBlock(/*MustBeDeclaration=*/false); 903 } 904 addUnwrappedLine(); 905 return; 906 case tok::objc_try: 907 // This branch isn't strictly necessary (the kw_try case below would 908 // do this too after the tok::at is parsed above). But be explicit. 909 parseTryCatch(); 910 return; 911 default: 912 break; 913 } 914 break; 915 case tok::kw_asm: 916 nextToken(); 917 if (FormatTok->is(tok::l_brace)) { 918 FormatTok->Type = TT_InlineASMBrace; 919 nextToken(); 920 while (FormatTok && FormatTok->isNot(tok::eof)) { 921 if (FormatTok->is(tok::r_brace)) { 922 FormatTok->Type = TT_InlineASMBrace; 923 nextToken(); 924 addUnwrappedLine(); 925 break; 926 } 927 FormatTok->Finalized = true; 928 nextToken(); 929 } 930 } 931 break; 932 case tok::kw_namespace: 933 parseNamespace(); 934 return; 935 case tok::kw_inline: 936 nextToken(); 937 if (FormatTok->Tok.is(tok::kw_namespace)) { 938 parseNamespace(); 939 return; 940 } 941 break; 942 case tok::kw_public: 943 case tok::kw_protected: 944 case tok::kw_private: 945 if (Style.Language == FormatStyle::LK_Java || 946 Style.Language == FormatStyle::LK_JavaScript) 947 nextToken(); 948 else 949 parseAccessSpecifier(); 950 return; 951 case tok::kw_if: 952 parseIfThenElse(); 953 return; 954 case tok::kw_for: 955 case tok::kw_while: 956 parseForOrWhileLoop(); 957 return; 958 case tok::kw_do: 959 parseDoWhile(); 960 return; 961 case tok::kw_switch: 962 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 963 // 'switch: string' field declaration. 964 break; 965 parseSwitch(); 966 return; 967 case tok::kw_default: 968 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 969 // 'default: string' field declaration. 970 break; 971 nextToken(); 972 parseLabel(); 973 return; 974 case tok::kw_case: 975 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 976 // 'case: string' field declaration. 977 break; 978 parseCaseLabel(); 979 return; 980 case tok::kw_try: 981 case tok::kw___try: 982 parseTryCatch(); 983 return; 984 case tok::kw_extern: 985 nextToken(); 986 if (FormatTok->Tok.is(tok::string_literal)) { 987 nextToken(); 988 if (FormatTok->Tok.is(tok::l_brace)) { 989 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 990 addUnwrappedLine(); 991 return; 992 } 993 } 994 break; 995 case tok::kw_export: 996 if (Style.Language == FormatStyle::LK_JavaScript) { 997 parseJavaScriptEs6ImportExport(); 998 return; 999 } 1000 break; 1001 case tok::identifier: 1002 if (FormatTok->is(TT_ForEachMacro)) { 1003 parseForOrWhileLoop(); 1004 return; 1005 } 1006 if (FormatTok->is(TT_MacroBlockBegin)) { 1007 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1008 /*MunchSemi=*/false); 1009 return; 1010 } 1011 if (FormatTok->is(Keywords.kw_import)) { 1012 if (Style.Language == FormatStyle::LK_JavaScript) { 1013 parseJavaScriptEs6ImportExport(); 1014 return; 1015 } 1016 if (Style.Language == FormatStyle::LK_Proto) { 1017 nextToken(); 1018 if (FormatTok->is(tok::kw_public)) 1019 nextToken(); 1020 if (!FormatTok->is(tok::string_literal)) 1021 return; 1022 nextToken(); 1023 if (FormatTok->is(tok::semi)) 1024 nextToken(); 1025 addUnwrappedLine(); 1026 return; 1027 } 1028 } 1029 if (Style.isCpp() && 1030 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1031 Keywords.kw_slots, Keywords.kw_qslots)) { 1032 nextToken(); 1033 if (FormatTok->is(tok::colon)) { 1034 nextToken(); 1035 addUnwrappedLine(); 1036 return; 1037 } 1038 } 1039 // In all other cases, parse the declaration. 1040 break; 1041 default: 1042 break; 1043 } 1044 do { 1045 const FormatToken *Previous = getPreviousToken(); 1046 switch (FormatTok->Tok.getKind()) { 1047 case tok::at: 1048 nextToken(); 1049 if (FormatTok->Tok.is(tok::l_brace)) { 1050 nextToken(); 1051 parseBracedList(); 1052 } 1053 break; 1054 case tok::kw_enum: 1055 // Ignore if this is part of "template <enum ...". 1056 if (Previous && Previous->is(tok::less)) { 1057 nextToken(); 1058 break; 1059 } 1060 1061 // parseEnum falls through and does not yet add an unwrapped line as an 1062 // enum definition can start a structural element. 1063 if (!parseEnum()) 1064 break; 1065 // This only applies for C++. 1066 if (!Style.isCpp()) { 1067 addUnwrappedLine(); 1068 return; 1069 } 1070 break; 1071 case tok::kw_typedef: 1072 nextToken(); 1073 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1074 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 1075 parseEnum(); 1076 break; 1077 case tok::kw_struct: 1078 case tok::kw_union: 1079 case tok::kw_class: 1080 // parseRecord falls through and does not yet add an unwrapped line as a 1081 // record declaration or definition can start a structural element. 1082 parseRecord(); 1083 // This does not apply for Java and JavaScript. 1084 if (Style.Language == FormatStyle::LK_Java || 1085 Style.Language == FormatStyle::LK_JavaScript) { 1086 if (FormatTok->is(tok::semi)) 1087 nextToken(); 1088 addUnwrappedLine(); 1089 return; 1090 } 1091 break; 1092 case tok::period: 1093 nextToken(); 1094 // In Java, classes have an implicit static member "class". 1095 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1096 FormatTok->is(tok::kw_class)) 1097 nextToken(); 1098 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1099 FormatTok->Tok.getIdentifierInfo()) 1100 // JavaScript only has pseudo keywords, all keywords are allowed to 1101 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1102 nextToken(); 1103 break; 1104 case tok::semi: 1105 nextToken(); 1106 addUnwrappedLine(); 1107 return; 1108 case tok::r_brace: 1109 addUnwrappedLine(); 1110 return; 1111 case tok::l_paren: 1112 parseParens(); 1113 break; 1114 case tok::kw_operator: 1115 nextToken(); 1116 if (FormatTok->isBinaryOperator()) 1117 nextToken(); 1118 break; 1119 case tok::caret: 1120 nextToken(); 1121 if (FormatTok->Tok.isAnyIdentifier() || 1122 FormatTok->isSimpleTypeSpecifier()) 1123 nextToken(); 1124 if (FormatTok->is(tok::l_paren)) 1125 parseParens(); 1126 if (FormatTok->is(tok::l_brace)) 1127 parseChildBlock(); 1128 break; 1129 case tok::l_brace: 1130 if (!tryToParseBracedList()) { 1131 // A block outside of parentheses must be the last part of a 1132 // structural element. 1133 // FIXME: Figure out cases where this is not true, and add projections 1134 // for them (the one we know is missing are lambdas). 1135 if (Style.BraceWrapping.AfterFunction) 1136 addUnwrappedLine(); 1137 FormatTok->Type = TT_FunctionLBrace; 1138 parseBlock(/*MustBeDeclaration=*/false); 1139 addUnwrappedLine(); 1140 return; 1141 } 1142 // Otherwise this was a braced init list, and the structural 1143 // element continues. 1144 break; 1145 case tok::kw_try: 1146 // We arrive here when parsing function-try blocks. 1147 parseTryCatch(); 1148 return; 1149 case tok::identifier: { 1150 if (FormatTok->is(TT_MacroBlockEnd)) { 1151 addUnwrappedLine(); 1152 return; 1153 } 1154 1155 // Function declarations (as opposed to function expressions) are parsed 1156 // on their own unwrapped line by continuing this loop. Function 1157 // expressions (functions that are not on their own line) must not create 1158 // a new unwrapped line, so they are special cased below. 1159 size_t TokenCount = Line->Tokens.size(); 1160 if (Style.Language == FormatStyle::LK_JavaScript && 1161 FormatTok->is(Keywords.kw_function) && 1162 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1163 Keywords.kw_async)))) { 1164 tryToParseJSFunction(); 1165 break; 1166 } 1167 if ((Style.Language == FormatStyle::LK_JavaScript || 1168 Style.Language == FormatStyle::LK_Java) && 1169 FormatTok->is(Keywords.kw_interface)) { 1170 if (Style.Language == FormatStyle::LK_JavaScript) { 1171 // In JavaScript/TypeScript, "interface" can be used as a standalone 1172 // identifier, e.g. in `var interface = 1;`. If "interface" is 1173 // followed by another identifier, it is very like to be an actual 1174 // interface declaration. 1175 unsigned StoredPosition = Tokens->getPosition(); 1176 FormatToken *Next = Tokens->getNextToken(); 1177 FormatTok = Tokens->setPosition(StoredPosition); 1178 if (Next && !mustBeJSIdent(Keywords, Next)) { 1179 nextToken(); 1180 break; 1181 } 1182 } 1183 parseRecord(); 1184 addUnwrappedLine(); 1185 return; 1186 } 1187 1188 // See if the following token should start a new unwrapped line. 1189 StringRef Text = FormatTok->TokenText; 1190 nextToken(); 1191 if (Line->Tokens.size() == 1 && 1192 // JS doesn't have macros, and within classes colons indicate fields, 1193 // not labels. 1194 Style.Language != FormatStyle::LK_JavaScript) { 1195 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1196 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1197 parseLabel(); 1198 return; 1199 } 1200 // Recognize function-like macro usages without trailing semicolon as 1201 // well as free-standing macros like Q_OBJECT. 1202 bool FunctionLike = FormatTok->is(tok::l_paren); 1203 if (FunctionLike) 1204 parseParens(); 1205 1206 bool FollowedByNewline = 1207 CommentsBeforeNextToken.empty() 1208 ? FormatTok->NewlinesBefore > 0 1209 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1210 1211 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1212 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1213 addUnwrappedLine(); 1214 return; 1215 } 1216 } 1217 break; 1218 } 1219 case tok::equal: 1220 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1221 // TT_JsFatArrow. The always start an expression or a child block if 1222 // followed by a curly. 1223 if (FormatTok->is(TT_JsFatArrow)) { 1224 nextToken(); 1225 if (FormatTok->is(tok::l_brace)) 1226 parseChildBlock(); 1227 break; 1228 } 1229 1230 nextToken(); 1231 if (FormatTok->Tok.is(tok::l_brace)) { 1232 nextToken(); 1233 parseBracedList(); 1234 } else if (Style.Language == FormatStyle::LK_Proto && 1235 FormatTok->Tok.is(tok::less)) { 1236 nextToken(); 1237 parseBracedList(/*ContinueOnSemicolons=*/false, 1238 /*ClosingBraceKind=*/tok::greater); 1239 } 1240 break; 1241 case tok::l_square: 1242 parseSquare(); 1243 break; 1244 case tok::kw_new: 1245 parseNew(); 1246 break; 1247 default: 1248 nextToken(); 1249 break; 1250 } 1251 } while (!eof()); 1252 } 1253 1254 bool UnwrappedLineParser::tryToParseLambda() { 1255 if (!Style.isCpp()) { 1256 nextToken(); 1257 return false; 1258 } 1259 const FormatToken* Previous = getPreviousToken(); 1260 if (Previous && 1261 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1262 tok::kw_delete) || 1263 Previous->closesScope() || Previous->isSimpleTypeSpecifier())) { 1264 nextToken(); 1265 return false; 1266 } 1267 assert(FormatTok->is(tok::l_square)); 1268 FormatToken &LSquare = *FormatTok; 1269 if (!tryToParseLambdaIntroducer()) 1270 return false; 1271 1272 while (FormatTok->isNot(tok::l_brace)) { 1273 if (FormatTok->isSimpleTypeSpecifier()) { 1274 nextToken(); 1275 continue; 1276 } 1277 switch (FormatTok->Tok.getKind()) { 1278 case tok::l_brace: 1279 break; 1280 case tok::l_paren: 1281 parseParens(); 1282 break; 1283 case tok::amp: 1284 case tok::star: 1285 case tok::kw_const: 1286 case tok::comma: 1287 case tok::less: 1288 case tok::greater: 1289 case tok::identifier: 1290 case tok::numeric_constant: 1291 case tok::coloncolon: 1292 case tok::kw_mutable: 1293 nextToken(); 1294 break; 1295 case tok::arrow: 1296 FormatTok->Type = TT_LambdaArrow; 1297 nextToken(); 1298 break; 1299 default: 1300 return true; 1301 } 1302 } 1303 LSquare.Type = TT_LambdaLSquare; 1304 parseChildBlock(); 1305 return true; 1306 } 1307 1308 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1309 nextToken(); 1310 if (FormatTok->is(tok::equal)) { 1311 nextToken(); 1312 if (FormatTok->is(tok::r_square)) { 1313 nextToken(); 1314 return true; 1315 } 1316 if (FormatTok->isNot(tok::comma)) 1317 return false; 1318 nextToken(); 1319 } else if (FormatTok->is(tok::amp)) { 1320 nextToken(); 1321 if (FormatTok->is(tok::r_square)) { 1322 nextToken(); 1323 return true; 1324 } 1325 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1326 return false; 1327 } 1328 if (FormatTok->is(tok::comma)) 1329 nextToken(); 1330 } else if (FormatTok->is(tok::r_square)) { 1331 nextToken(); 1332 return true; 1333 } 1334 do { 1335 if (FormatTok->is(tok::amp)) 1336 nextToken(); 1337 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1338 return false; 1339 nextToken(); 1340 if (FormatTok->is(tok::ellipsis)) 1341 nextToken(); 1342 if (FormatTok->is(tok::comma)) { 1343 nextToken(); 1344 } else if (FormatTok->is(tok::r_square)) { 1345 nextToken(); 1346 return true; 1347 } else { 1348 return false; 1349 } 1350 } while (!eof()); 1351 return false; 1352 } 1353 1354 void UnwrappedLineParser::tryToParseJSFunction() { 1355 assert(FormatTok->is(Keywords.kw_function) || 1356 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1357 if (FormatTok->is(Keywords.kw_async)) 1358 nextToken(); 1359 // Consume "function". 1360 nextToken(); 1361 1362 // Consume * (generator function). Treat it like C++'s overloaded operators. 1363 if (FormatTok->is(tok::star)) { 1364 FormatTok->Type = TT_OverloadedOperator; 1365 nextToken(); 1366 } 1367 1368 // Consume function name. 1369 if (FormatTok->is(tok::identifier)) 1370 nextToken(); 1371 1372 if (FormatTok->isNot(tok::l_paren)) 1373 return; 1374 1375 // Parse formal parameter list. 1376 parseParens(); 1377 1378 if (FormatTok->is(tok::colon)) { 1379 // Parse a type definition. 1380 nextToken(); 1381 1382 // Eat the type declaration. For braced inline object types, balance braces, 1383 // otherwise just parse until finding an l_brace for the function body. 1384 if (FormatTok->is(tok::l_brace)) 1385 tryToParseBracedList(); 1386 else 1387 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1388 nextToken(); 1389 } 1390 1391 if (FormatTok->is(tok::semi)) 1392 return; 1393 1394 parseChildBlock(); 1395 } 1396 1397 bool UnwrappedLineParser::tryToParseBracedList() { 1398 if (FormatTok->BlockKind == BK_Unknown) 1399 calculateBraceTypes(); 1400 assert(FormatTok->BlockKind != BK_Unknown); 1401 if (FormatTok->BlockKind == BK_Block) 1402 return false; 1403 nextToken(); 1404 parseBracedList(); 1405 return true; 1406 } 1407 1408 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1409 tok::TokenKind ClosingBraceKind) { 1410 bool HasError = false; 1411 1412 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1413 // replace this by using parseAssigmentExpression() inside. 1414 do { 1415 if (Style.Language == FormatStyle::LK_JavaScript) { 1416 if (FormatTok->is(Keywords.kw_function) || 1417 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1418 tryToParseJSFunction(); 1419 continue; 1420 } 1421 if (FormatTok->is(TT_JsFatArrow)) { 1422 nextToken(); 1423 // Fat arrows can be followed by simple expressions or by child blocks 1424 // in curly braces. 1425 if (FormatTok->is(tok::l_brace)) { 1426 parseChildBlock(); 1427 continue; 1428 } 1429 } 1430 if (FormatTok->is(tok::l_brace)) { 1431 // Could be a method inside of a braced list `{a() { return 1; }}`. 1432 if (tryToParseBracedList()) 1433 continue; 1434 parseChildBlock(); 1435 } 1436 } 1437 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1438 nextToken(); 1439 return !HasError; 1440 } 1441 switch (FormatTok->Tok.getKind()) { 1442 case tok::caret: 1443 nextToken(); 1444 if (FormatTok->is(tok::l_brace)) { 1445 parseChildBlock(); 1446 } 1447 break; 1448 case tok::l_square: 1449 tryToParseLambda(); 1450 break; 1451 case tok::l_paren: 1452 parseParens(); 1453 // JavaScript can just have free standing methods and getters/setters in 1454 // object literals. Detect them by a "{" following ")". 1455 if (Style.Language == FormatStyle::LK_JavaScript) { 1456 if (FormatTok->is(tok::l_brace)) 1457 parseChildBlock(); 1458 break; 1459 } 1460 break; 1461 case tok::l_brace: 1462 // Assume there are no blocks inside a braced init list apart 1463 // from the ones we explicitly parse out (like lambdas). 1464 FormatTok->BlockKind = BK_BracedInit; 1465 nextToken(); 1466 parseBracedList(); 1467 break; 1468 case tok::less: 1469 if (Style.Language == FormatStyle::LK_Proto) { 1470 nextToken(); 1471 parseBracedList(/*ContinueOnSemicolons=*/false, 1472 /*ClosingBraceKind=*/tok::greater); 1473 } else { 1474 nextToken(); 1475 } 1476 break; 1477 case tok::semi: 1478 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1479 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1480 // used for error recovery if we have otherwise determined that this is 1481 // a braced list. 1482 if (Style.Language == FormatStyle::LK_JavaScript) { 1483 nextToken(); 1484 break; 1485 } 1486 HasError = true; 1487 if (!ContinueOnSemicolons) 1488 return !HasError; 1489 nextToken(); 1490 break; 1491 case tok::comma: 1492 nextToken(); 1493 break; 1494 default: 1495 nextToken(); 1496 break; 1497 } 1498 } while (!eof()); 1499 return false; 1500 } 1501 1502 void UnwrappedLineParser::parseParens() { 1503 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1504 nextToken(); 1505 do { 1506 switch (FormatTok->Tok.getKind()) { 1507 case tok::l_paren: 1508 parseParens(); 1509 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1510 parseChildBlock(); 1511 break; 1512 case tok::r_paren: 1513 nextToken(); 1514 return; 1515 case tok::r_brace: 1516 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1517 return; 1518 case tok::l_square: 1519 tryToParseLambda(); 1520 break; 1521 case tok::l_brace: 1522 if (!tryToParseBracedList()) 1523 parseChildBlock(); 1524 break; 1525 case tok::at: 1526 nextToken(); 1527 if (FormatTok->Tok.is(tok::l_brace)) { 1528 nextToken(); 1529 parseBracedList(); 1530 } 1531 break; 1532 case tok::kw_class: 1533 if (Style.Language == FormatStyle::LK_JavaScript) 1534 parseRecord(/*ParseAsExpr=*/true); 1535 else 1536 nextToken(); 1537 break; 1538 case tok::identifier: 1539 if (Style.Language == FormatStyle::LK_JavaScript && 1540 (FormatTok->is(Keywords.kw_function) || 1541 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1542 tryToParseJSFunction(); 1543 else 1544 nextToken(); 1545 break; 1546 default: 1547 nextToken(); 1548 break; 1549 } 1550 } while (!eof()); 1551 } 1552 1553 void UnwrappedLineParser::parseSquare() { 1554 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1555 if (tryToParseLambda()) 1556 return; 1557 do { 1558 switch (FormatTok->Tok.getKind()) { 1559 case tok::l_paren: 1560 parseParens(); 1561 break; 1562 case tok::r_square: 1563 nextToken(); 1564 return; 1565 case tok::r_brace: 1566 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1567 return; 1568 case tok::l_square: 1569 parseSquare(); 1570 break; 1571 case tok::l_brace: { 1572 if (!tryToParseBracedList()) 1573 parseChildBlock(); 1574 break; 1575 } 1576 case tok::at: 1577 nextToken(); 1578 if (FormatTok->Tok.is(tok::l_brace)) { 1579 nextToken(); 1580 parseBracedList(); 1581 } 1582 break; 1583 default: 1584 nextToken(); 1585 break; 1586 } 1587 } while (!eof()); 1588 } 1589 1590 void UnwrappedLineParser::parseIfThenElse() { 1591 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1592 nextToken(); 1593 if (FormatTok->Tok.is(tok::kw_constexpr)) 1594 nextToken(); 1595 if (FormatTok->Tok.is(tok::l_paren)) 1596 parseParens(); 1597 bool NeedsUnwrappedLine = false; 1598 if (FormatTok->Tok.is(tok::l_brace)) { 1599 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1600 parseBlock(/*MustBeDeclaration=*/false); 1601 if (Style.BraceWrapping.BeforeElse) 1602 addUnwrappedLine(); 1603 else 1604 NeedsUnwrappedLine = true; 1605 } else { 1606 addUnwrappedLine(); 1607 ++Line->Level; 1608 parseStructuralElement(); 1609 --Line->Level; 1610 } 1611 if (FormatTok->Tok.is(tok::kw_else)) { 1612 nextToken(); 1613 if (FormatTok->Tok.is(tok::l_brace)) { 1614 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1615 parseBlock(/*MustBeDeclaration=*/false); 1616 addUnwrappedLine(); 1617 } else if (FormatTok->Tok.is(tok::kw_if)) { 1618 parseIfThenElse(); 1619 } else { 1620 addUnwrappedLine(); 1621 ++Line->Level; 1622 parseStructuralElement(); 1623 if (FormatTok->is(tok::eof)) 1624 addUnwrappedLine(); 1625 --Line->Level; 1626 } 1627 } else if (NeedsUnwrappedLine) { 1628 addUnwrappedLine(); 1629 } 1630 } 1631 1632 void UnwrappedLineParser::parseTryCatch() { 1633 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1634 nextToken(); 1635 bool NeedsUnwrappedLine = false; 1636 if (FormatTok->is(tok::colon)) { 1637 // We are in a function try block, what comes is an initializer list. 1638 nextToken(); 1639 while (FormatTok->is(tok::identifier)) { 1640 nextToken(); 1641 if (FormatTok->is(tok::l_paren)) 1642 parseParens(); 1643 if (FormatTok->is(tok::comma)) 1644 nextToken(); 1645 } 1646 } 1647 // Parse try with resource. 1648 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1649 parseParens(); 1650 } 1651 if (FormatTok->is(tok::l_brace)) { 1652 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1653 parseBlock(/*MustBeDeclaration=*/false); 1654 if (Style.BraceWrapping.BeforeCatch) { 1655 addUnwrappedLine(); 1656 } else { 1657 NeedsUnwrappedLine = true; 1658 } 1659 } else if (!FormatTok->is(tok::kw_catch)) { 1660 // The C++ standard requires a compound-statement after a try. 1661 // If there's none, we try to assume there's a structuralElement 1662 // and try to continue. 1663 addUnwrappedLine(); 1664 ++Line->Level; 1665 parseStructuralElement(); 1666 --Line->Level; 1667 } 1668 while (1) { 1669 if (FormatTok->is(tok::at)) 1670 nextToken(); 1671 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1672 tok::kw___finally) || 1673 ((Style.Language == FormatStyle::LK_Java || 1674 Style.Language == FormatStyle::LK_JavaScript) && 1675 FormatTok->is(Keywords.kw_finally)) || 1676 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1677 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1678 break; 1679 nextToken(); 1680 while (FormatTok->isNot(tok::l_brace)) { 1681 if (FormatTok->is(tok::l_paren)) { 1682 parseParens(); 1683 continue; 1684 } 1685 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1686 return; 1687 nextToken(); 1688 } 1689 NeedsUnwrappedLine = false; 1690 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1691 parseBlock(/*MustBeDeclaration=*/false); 1692 if (Style.BraceWrapping.BeforeCatch) 1693 addUnwrappedLine(); 1694 else 1695 NeedsUnwrappedLine = true; 1696 } 1697 if (NeedsUnwrappedLine) 1698 addUnwrappedLine(); 1699 } 1700 1701 void UnwrappedLineParser::parseNamespace() { 1702 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1703 1704 const FormatToken &InitialToken = *FormatTok; 1705 nextToken(); 1706 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1707 nextToken(); 1708 if (FormatTok->Tok.is(tok::l_brace)) { 1709 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1710 addUnwrappedLine(); 1711 1712 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1713 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1714 DeclarationScopeStack.size() > 1); 1715 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1716 // Munch the semicolon after a namespace. This is more common than one would 1717 // think. Puttin the semicolon into its own line is very ugly. 1718 if (FormatTok->Tok.is(tok::semi)) 1719 nextToken(); 1720 addUnwrappedLine(); 1721 } 1722 // FIXME: Add error handling. 1723 } 1724 1725 void UnwrappedLineParser::parseNew() { 1726 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1727 nextToken(); 1728 if (Style.Language != FormatStyle::LK_Java) 1729 return; 1730 1731 // In Java, we can parse everything up to the parens, which aren't optional. 1732 do { 1733 // There should not be a ;, { or } before the new's open paren. 1734 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1735 return; 1736 1737 // Consume the parens. 1738 if (FormatTok->is(tok::l_paren)) { 1739 parseParens(); 1740 1741 // If there is a class body of an anonymous class, consume that as child. 1742 if (FormatTok->is(tok::l_brace)) 1743 parseChildBlock(); 1744 return; 1745 } 1746 nextToken(); 1747 } while (!eof()); 1748 } 1749 1750 void UnwrappedLineParser::parseForOrWhileLoop() { 1751 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1752 "'for', 'while' or foreach macro expected"); 1753 nextToken(); 1754 // JS' for await ( ... 1755 if (Style.Language == FormatStyle::LK_JavaScript && 1756 FormatTok->is(Keywords.kw_await)) 1757 nextToken(); 1758 if (FormatTok->Tok.is(tok::l_paren)) 1759 parseParens(); 1760 if (FormatTok->Tok.is(tok::l_brace)) { 1761 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1762 parseBlock(/*MustBeDeclaration=*/false); 1763 addUnwrappedLine(); 1764 } else { 1765 addUnwrappedLine(); 1766 ++Line->Level; 1767 parseStructuralElement(); 1768 --Line->Level; 1769 } 1770 } 1771 1772 void UnwrappedLineParser::parseDoWhile() { 1773 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1774 nextToken(); 1775 if (FormatTok->Tok.is(tok::l_brace)) { 1776 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1777 parseBlock(/*MustBeDeclaration=*/false); 1778 if (Style.BraceWrapping.IndentBraces) 1779 addUnwrappedLine(); 1780 } else { 1781 addUnwrappedLine(); 1782 ++Line->Level; 1783 parseStructuralElement(); 1784 --Line->Level; 1785 } 1786 1787 // FIXME: Add error handling. 1788 if (!FormatTok->Tok.is(tok::kw_while)) { 1789 addUnwrappedLine(); 1790 return; 1791 } 1792 1793 nextToken(); 1794 parseStructuralElement(); 1795 } 1796 1797 void UnwrappedLineParser::parseLabel() { 1798 nextToken(); 1799 unsigned OldLineLevel = Line->Level; 1800 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1801 --Line->Level; 1802 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1803 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1804 parseBlock(/*MustBeDeclaration=*/false); 1805 if (FormatTok->Tok.is(tok::kw_break)) { 1806 if (Style.BraceWrapping.AfterControlStatement) 1807 addUnwrappedLine(); 1808 parseStructuralElement(); 1809 } 1810 addUnwrappedLine(); 1811 } else { 1812 if (FormatTok->is(tok::semi)) 1813 nextToken(); 1814 addUnwrappedLine(); 1815 } 1816 Line->Level = OldLineLevel; 1817 if (FormatTok->isNot(tok::l_brace)) { 1818 parseStructuralElement(); 1819 addUnwrappedLine(); 1820 } 1821 } 1822 1823 void UnwrappedLineParser::parseCaseLabel() { 1824 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1825 // FIXME: fix handling of complex expressions here. 1826 do { 1827 nextToken(); 1828 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1829 parseLabel(); 1830 } 1831 1832 void UnwrappedLineParser::parseSwitch() { 1833 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1834 nextToken(); 1835 if (FormatTok->Tok.is(tok::l_paren)) 1836 parseParens(); 1837 if (FormatTok->Tok.is(tok::l_brace)) { 1838 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1839 parseBlock(/*MustBeDeclaration=*/false); 1840 addUnwrappedLine(); 1841 } else { 1842 addUnwrappedLine(); 1843 ++Line->Level; 1844 parseStructuralElement(); 1845 --Line->Level; 1846 } 1847 } 1848 1849 void UnwrappedLineParser::parseAccessSpecifier() { 1850 nextToken(); 1851 // Understand Qt's slots. 1852 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1853 nextToken(); 1854 // Otherwise, we don't know what it is, and we'd better keep the next token. 1855 if (FormatTok->Tok.is(tok::colon)) 1856 nextToken(); 1857 addUnwrappedLine(); 1858 } 1859 1860 bool UnwrappedLineParser::parseEnum() { 1861 // Won't be 'enum' for NS_ENUMs. 1862 if (FormatTok->Tok.is(tok::kw_enum)) 1863 nextToken(); 1864 1865 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1866 // declarations. An "enum" keyword followed by a colon would be a syntax 1867 // error and thus assume it is just an identifier. 1868 if (Style.Language == FormatStyle::LK_JavaScript && 1869 FormatTok->isOneOf(tok::colon, tok::question)) 1870 return false; 1871 1872 // Eat up enum class ... 1873 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1874 nextToken(); 1875 1876 while (FormatTok->Tok.getIdentifierInfo() || 1877 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1878 tok::greater, tok::comma, tok::question)) { 1879 nextToken(); 1880 // We can have macros or attributes in between 'enum' and the enum name. 1881 if (FormatTok->is(tok::l_paren)) 1882 parseParens(); 1883 if (FormatTok->is(tok::identifier)) { 1884 nextToken(); 1885 // If there are two identifiers in a row, this is likely an elaborate 1886 // return type. In Java, this can be "implements", etc. 1887 if (Style.isCpp() && FormatTok->is(tok::identifier)) 1888 return false; 1889 } 1890 } 1891 1892 // Just a declaration or something is wrong. 1893 if (FormatTok->isNot(tok::l_brace)) 1894 return true; 1895 FormatTok->BlockKind = BK_Block; 1896 1897 if (Style.Language == FormatStyle::LK_Java) { 1898 // Java enums are different. 1899 parseJavaEnumBody(); 1900 return true; 1901 } 1902 if (Style.Language == FormatStyle::LK_Proto) { 1903 parseBlock(/*MustBeDeclaration=*/true); 1904 return true; 1905 } 1906 1907 // Parse enum body. 1908 nextToken(); 1909 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1910 if (HasError) { 1911 if (FormatTok->is(tok::semi)) 1912 nextToken(); 1913 addUnwrappedLine(); 1914 } 1915 return true; 1916 1917 // There is no addUnwrappedLine() here so that we fall through to parsing a 1918 // structural element afterwards. Thus, in "enum A {} n, m;", 1919 // "} n, m;" will end up in one unwrapped line. 1920 } 1921 1922 void UnwrappedLineParser::parseJavaEnumBody() { 1923 // Determine whether the enum is simple, i.e. does not have a semicolon or 1924 // constants with class bodies. Simple enums can be formatted like braced 1925 // lists, contracted to a single line, etc. 1926 unsigned StoredPosition = Tokens->getPosition(); 1927 bool IsSimple = true; 1928 FormatToken *Tok = Tokens->getNextToken(); 1929 while (Tok) { 1930 if (Tok->is(tok::r_brace)) 1931 break; 1932 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1933 IsSimple = false; 1934 break; 1935 } 1936 // FIXME: This will also mark enums with braces in the arguments to enum 1937 // constants as "not simple". This is probably fine in practice, though. 1938 Tok = Tokens->getNextToken(); 1939 } 1940 FormatTok = Tokens->setPosition(StoredPosition); 1941 1942 if (IsSimple) { 1943 nextToken(); 1944 parseBracedList(); 1945 addUnwrappedLine(); 1946 return; 1947 } 1948 1949 // Parse the body of a more complex enum. 1950 // First add a line for everything up to the "{". 1951 nextToken(); 1952 addUnwrappedLine(); 1953 ++Line->Level; 1954 1955 // Parse the enum constants. 1956 while (FormatTok) { 1957 if (FormatTok->is(tok::l_brace)) { 1958 // Parse the constant's class body. 1959 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1960 /*MunchSemi=*/false); 1961 } else if (FormatTok->is(tok::l_paren)) { 1962 parseParens(); 1963 } else if (FormatTok->is(tok::comma)) { 1964 nextToken(); 1965 addUnwrappedLine(); 1966 } else if (FormatTok->is(tok::semi)) { 1967 nextToken(); 1968 addUnwrappedLine(); 1969 break; 1970 } else if (FormatTok->is(tok::r_brace)) { 1971 addUnwrappedLine(); 1972 break; 1973 } else { 1974 nextToken(); 1975 } 1976 } 1977 1978 // Parse the class body after the enum's ";" if any. 1979 parseLevel(/*HasOpeningBrace=*/true); 1980 nextToken(); 1981 --Line->Level; 1982 addUnwrappedLine(); 1983 } 1984 1985 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 1986 const FormatToken &InitialToken = *FormatTok; 1987 nextToken(); 1988 1989 // The actual identifier can be a nested name specifier, and in macros 1990 // it is often token-pasted. 1991 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1992 tok::kw___attribute, tok::kw___declspec, 1993 tok::kw_alignas) || 1994 ((Style.Language == FormatStyle::LK_Java || 1995 Style.Language == FormatStyle::LK_JavaScript) && 1996 FormatTok->isOneOf(tok::period, tok::comma))) { 1997 if (Style.Language == FormatStyle::LK_JavaScript && 1998 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 1999 // JavaScript/TypeScript supports inline object types in 2000 // extends/implements positions: 2001 // class Foo implements {bar: number} { } 2002 nextToken(); 2003 if (FormatTok->is(tok::l_brace)) { 2004 tryToParseBracedList(); 2005 continue; 2006 } 2007 } 2008 bool IsNonMacroIdentifier = 2009 FormatTok->is(tok::identifier) && 2010 FormatTok->TokenText != FormatTok->TokenText.upper(); 2011 nextToken(); 2012 // We can have macros or attributes in between 'class' and the class name. 2013 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 2014 parseParens(); 2015 } 2016 2017 // Note that parsing away template declarations here leads to incorrectly 2018 // accepting function declarations as record declarations. 2019 // In general, we cannot solve this problem. Consider: 2020 // class A<int> B() {} 2021 // which can be a function definition or a class definition when B() is a 2022 // macro. If we find enough real-world cases where this is a problem, we 2023 // can parse for the 'template' keyword in the beginning of the statement, 2024 // and thus rule out the record production in case there is no template 2025 // (this would still leave us with an ambiguity between template function 2026 // and class declarations). 2027 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2028 while (!eof()) { 2029 if (FormatTok->is(tok::l_brace)) { 2030 calculateBraceTypes(/*ExpectClassBody=*/true); 2031 if (!tryToParseBracedList()) 2032 break; 2033 } 2034 if (FormatTok->Tok.is(tok::semi)) 2035 return; 2036 nextToken(); 2037 } 2038 } 2039 if (FormatTok->Tok.is(tok::l_brace)) { 2040 if (ParseAsExpr) { 2041 parseChildBlock(); 2042 } else { 2043 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2044 addUnwrappedLine(); 2045 2046 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2047 /*MunchSemi=*/false); 2048 } 2049 } 2050 // There is no addUnwrappedLine() here so that we fall through to parsing a 2051 // structural element afterwards. Thus, in "class A {} n, m;", 2052 // "} n, m;" will end up in one unwrapped line. 2053 } 2054 2055 void UnwrappedLineParser::parseObjCProtocolList() { 2056 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2057 do 2058 nextToken(); 2059 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2060 nextToken(); // Skip '>'. 2061 } 2062 2063 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2064 do { 2065 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2066 nextToken(); 2067 addUnwrappedLine(); 2068 break; 2069 } 2070 if (FormatTok->is(tok::l_brace)) { 2071 parseBlock(/*MustBeDeclaration=*/false); 2072 // In ObjC interfaces, nothing should be following the "}". 2073 addUnwrappedLine(); 2074 } else if (FormatTok->is(tok::r_brace)) { 2075 // Ignore stray "}". parseStructuralElement doesn't consume them. 2076 nextToken(); 2077 addUnwrappedLine(); 2078 } else { 2079 parseStructuralElement(); 2080 } 2081 } while (!eof()); 2082 } 2083 2084 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2085 nextToken(); 2086 nextToken(); // interface name 2087 2088 // @interface can be followed by either a base class, or a category. 2089 if (FormatTok->Tok.is(tok::colon)) { 2090 nextToken(); 2091 nextToken(); // base class name 2092 } else if (FormatTok->Tok.is(tok::l_paren)) 2093 // Skip category, if present. 2094 parseParens(); 2095 2096 if (FormatTok->Tok.is(tok::less)) 2097 parseObjCProtocolList(); 2098 2099 if (FormatTok->Tok.is(tok::l_brace)) { 2100 if (Style.BraceWrapping.AfterObjCDeclaration) 2101 addUnwrappedLine(); 2102 parseBlock(/*MustBeDeclaration=*/true); 2103 } 2104 2105 // With instance variables, this puts '}' on its own line. Without instance 2106 // variables, this ends the @interface line. 2107 addUnwrappedLine(); 2108 2109 parseObjCUntilAtEnd(); 2110 } 2111 2112 void UnwrappedLineParser::parseObjCProtocol() { 2113 nextToken(); 2114 nextToken(); // protocol name 2115 2116 if (FormatTok->Tok.is(tok::less)) 2117 parseObjCProtocolList(); 2118 2119 // Check for protocol declaration. 2120 if (FormatTok->Tok.is(tok::semi)) { 2121 nextToken(); 2122 return addUnwrappedLine(); 2123 } 2124 2125 addUnwrappedLine(); 2126 parseObjCUntilAtEnd(); 2127 } 2128 2129 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2130 bool IsImport = FormatTok->is(Keywords.kw_import); 2131 assert(IsImport || FormatTok->is(tok::kw_export)); 2132 nextToken(); 2133 2134 // Consume the "default" in "export default class/function". 2135 if (FormatTok->is(tok::kw_default)) 2136 nextToken(); 2137 2138 // Consume "async function", "function" and "default function", so that these 2139 // get parsed as free-standing JS functions, i.e. do not require a trailing 2140 // semicolon. 2141 if (FormatTok->is(Keywords.kw_async)) 2142 nextToken(); 2143 if (FormatTok->is(Keywords.kw_function)) { 2144 nextToken(); 2145 return; 2146 } 2147 2148 // For imports, `export *`, `export {...}`, consume the rest of the line up 2149 // to the terminating `;`. For everything else, just return and continue 2150 // parsing the structural element, i.e. the declaration or expression for 2151 // `export default`. 2152 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2153 !FormatTok->isStringLiteral()) 2154 return; 2155 2156 while (!eof()) { 2157 if (FormatTok->is(tok::semi)) 2158 return; 2159 if (Line->Tokens.size() == 0) { 2160 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2161 // import statement should terminate. 2162 return; 2163 } 2164 if (FormatTok->is(tok::l_brace)) { 2165 FormatTok->BlockKind = BK_Block; 2166 nextToken(); 2167 parseBracedList(); 2168 } else { 2169 nextToken(); 2170 } 2171 } 2172 } 2173 2174 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2175 StringRef Prefix = "") { 2176 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 2177 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2178 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2179 E = Line.Tokens.end(); 2180 I != E; ++I) { 2181 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2182 << "T=" << I->Tok->Type 2183 << ", OC=" << I->Tok->OriginalColumn << "] "; 2184 } 2185 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2186 E = Line.Tokens.end(); 2187 I != E; ++I) { 2188 const UnwrappedLineNode &Node = *I; 2189 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2190 I = Node.Children.begin(), 2191 E = Node.Children.end(); 2192 I != E; ++I) { 2193 printDebugInfo(*I, "\nChild: "); 2194 } 2195 } 2196 llvm::dbgs() << "\n"; 2197 } 2198 2199 void UnwrappedLineParser::addUnwrappedLine() { 2200 if (Line->Tokens.empty()) 2201 return; 2202 DEBUG({ 2203 if (CurrentLines == &Lines) 2204 printDebugInfo(*Line); 2205 }); 2206 CurrentLines->push_back(std::move(*Line)); 2207 Line->Tokens.clear(); 2208 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2209 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2210 CurrentLines->append( 2211 std::make_move_iterator(PreprocessorDirectives.begin()), 2212 std::make_move_iterator(PreprocessorDirectives.end())); 2213 PreprocessorDirectives.clear(); 2214 } 2215 } 2216 2217 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2218 2219 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2220 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2221 FormatTok.NewlinesBefore > 0; 2222 } 2223 2224 // Checks if \p FormatTok is a line comment that continues the line comment 2225 // section on \p Line. 2226 static bool continuesLineCommentSection(const FormatToken &FormatTok, 2227 const UnwrappedLine &Line, 2228 llvm::Regex &CommentPragmasRegex) { 2229 if (Line.Tokens.empty()) 2230 return false; 2231 2232 StringRef IndentContent = FormatTok.TokenText; 2233 if (FormatTok.TokenText.startswith("//") || 2234 FormatTok.TokenText.startswith("/*")) 2235 IndentContent = FormatTok.TokenText.substr(2); 2236 if (CommentPragmasRegex.match(IndentContent)) 2237 return false; 2238 2239 // If Line starts with a line comment, then FormatTok continues the comment 2240 // section if its original column is greater or equal to the original start 2241 // column of the line. 2242 // 2243 // Define the min column token of a line as follows: if a line ends in '{' or 2244 // contains a '{' followed by a line comment, then the min column token is 2245 // that '{'. Otherwise, the min column token of the line is the first token of 2246 // the line. 2247 // 2248 // If Line starts with a token other than a line comment, then FormatTok 2249 // continues the comment section if its original column is greater than the 2250 // original start column of the min column token of the line. 2251 // 2252 // For example, the second line comment continues the first in these cases: 2253 // 2254 // // first line 2255 // // second line 2256 // 2257 // and: 2258 // 2259 // // first line 2260 // // second line 2261 // 2262 // and: 2263 // 2264 // int i; // first line 2265 // // second line 2266 // 2267 // and: 2268 // 2269 // do { // first line 2270 // // second line 2271 // int i; 2272 // } while (true); 2273 // 2274 // and: 2275 // 2276 // enum { 2277 // a, // first line 2278 // // second line 2279 // b 2280 // }; 2281 // 2282 // The second line comment doesn't continue the first in these cases: 2283 // 2284 // // first line 2285 // // second line 2286 // 2287 // and: 2288 // 2289 // int i; // first line 2290 // // second line 2291 // 2292 // and: 2293 // 2294 // do { // first line 2295 // // second line 2296 // int i; 2297 // } while (true); 2298 // 2299 // and: 2300 // 2301 // enum { 2302 // a, // first line 2303 // // second line 2304 // }; 2305 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2306 2307 // Scan for '{//'. If found, use the column of '{' as a min column for line 2308 // comment section continuation. 2309 const FormatToken *PreviousToken = nullptr; 2310 for (const UnwrappedLineNode &Node : Line.Tokens) { 2311 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2312 isLineComment(*Node.Tok)) { 2313 MinColumnToken = PreviousToken; 2314 break; 2315 } 2316 PreviousToken = Node.Tok; 2317 2318 // Grab the last newline preceding a token in this unwrapped line. 2319 if (Node.Tok->NewlinesBefore > 0) { 2320 MinColumnToken = Node.Tok; 2321 } 2322 } 2323 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2324 MinColumnToken = PreviousToken; 2325 } 2326 2327 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2328 MinColumnToken); 2329 } 2330 2331 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2332 bool JustComments = Line->Tokens.empty(); 2333 for (SmallVectorImpl<FormatToken *>::const_iterator 2334 I = CommentsBeforeNextToken.begin(), 2335 E = CommentsBeforeNextToken.end(); 2336 I != E; ++I) { 2337 // Line comments that belong to the same line comment section are put on the 2338 // same line since later we might want to reflow content between them. 2339 // Additional fine-grained breaking of line comment sections is controlled 2340 // by the class BreakableLineCommentSection in case it is desirable to keep 2341 // several line comment sections in the same unwrapped line. 2342 // 2343 // FIXME: Consider putting separate line comment sections as children to the 2344 // unwrapped line instead. 2345 (*I)->ContinuesLineCommentSection = 2346 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2347 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2348 addUnwrappedLine(); 2349 pushToken(*I); 2350 } 2351 if (NewlineBeforeNext && JustComments) 2352 addUnwrappedLine(); 2353 CommentsBeforeNextToken.clear(); 2354 } 2355 2356 void UnwrappedLineParser::nextToken(int LevelDifference) { 2357 if (eof()) 2358 return; 2359 flushComments(isOnNewLine(*FormatTok)); 2360 pushToken(FormatTok); 2361 if (Style.Language != FormatStyle::LK_JavaScript) 2362 readToken(LevelDifference); 2363 else 2364 readTokenWithJavaScriptASI(); 2365 } 2366 2367 const FormatToken *UnwrappedLineParser::getPreviousToken() { 2368 // FIXME: This is a dirty way to access the previous token. Find a better 2369 // solution. 2370 if (!Line || Line->Tokens.empty()) 2371 return nullptr; 2372 return Line->Tokens.back().Tok; 2373 } 2374 2375 void UnwrappedLineParser::distributeComments( 2376 const SmallVectorImpl<FormatToken *> &Comments, 2377 const FormatToken *NextTok) { 2378 // Whether or not a line comment token continues a line is controlled by 2379 // the method continuesLineCommentSection, with the following caveat: 2380 // 2381 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2382 // that each comment line from the trail is aligned with the next token, if 2383 // the next token exists. If a trail exists, the beginning of the maximal 2384 // trail is marked as a start of a new comment section. 2385 // 2386 // For example in this code: 2387 // 2388 // int a; // line about a 2389 // // line 1 about b 2390 // // line 2 about b 2391 // int b; 2392 // 2393 // the two lines about b form a maximal trail, so there are two sections, the 2394 // first one consisting of the single comment "// line about a" and the 2395 // second one consisting of the next two comments. 2396 if (Comments.empty()) 2397 return; 2398 bool ShouldPushCommentsInCurrentLine = true; 2399 bool HasTrailAlignedWithNextToken = false; 2400 unsigned StartOfTrailAlignedWithNextToken = 0; 2401 if (NextTok) { 2402 // We are skipping the first element intentionally. 2403 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2404 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2405 HasTrailAlignedWithNextToken = true; 2406 StartOfTrailAlignedWithNextToken = i; 2407 } 2408 } 2409 } 2410 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2411 FormatToken *FormatTok = Comments[i]; 2412 if (HasTrailAlignedWithNextToken && 2413 i == StartOfTrailAlignedWithNextToken) { 2414 FormatTok->ContinuesLineCommentSection = false; 2415 } else { 2416 FormatTok->ContinuesLineCommentSection = 2417 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2418 } 2419 if (!FormatTok->ContinuesLineCommentSection && 2420 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2421 ShouldPushCommentsInCurrentLine = false; 2422 } 2423 if (ShouldPushCommentsInCurrentLine) { 2424 pushToken(FormatTok); 2425 } else { 2426 CommentsBeforeNextToken.push_back(FormatTok); 2427 } 2428 } 2429 } 2430 2431 void UnwrappedLineParser::readToken(int LevelDifference) { 2432 SmallVector<FormatToken *, 1> Comments; 2433 do { 2434 FormatTok = Tokens->getNextToken(); 2435 assert(FormatTok); 2436 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2437 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2438 distributeComments(Comments, FormatTok); 2439 Comments.clear(); 2440 // If there is an unfinished unwrapped line, we flush the preprocessor 2441 // directives only after that unwrapped line was finished later. 2442 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2443 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2444 assert((LevelDifference >= 0 || 2445 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 2446 "LevelDifference makes Line->Level negative"); 2447 Line->Level += LevelDifference; 2448 // Comments stored before the preprocessor directive need to be output 2449 // before the preprocessor directive, at the same level as the 2450 // preprocessor directive, as we consider them to apply to the directive. 2451 flushComments(isOnNewLine(*FormatTok)); 2452 parsePPDirective(); 2453 } 2454 while (FormatTok->Type == TT_ConflictStart || 2455 FormatTok->Type == TT_ConflictEnd || 2456 FormatTok->Type == TT_ConflictAlternative) { 2457 if (FormatTok->Type == TT_ConflictStart) { 2458 conditionalCompilationStart(/*Unreachable=*/false); 2459 } else if (FormatTok->Type == TT_ConflictAlternative) { 2460 conditionalCompilationAlternative(); 2461 } else if (FormatTok->Type == TT_ConflictEnd) { 2462 conditionalCompilationEnd(); 2463 } 2464 FormatTok = Tokens->getNextToken(); 2465 FormatTok->MustBreakBefore = true; 2466 } 2467 2468 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 2469 !Line->InPPDirective) { 2470 continue; 2471 } 2472 2473 if (!FormatTok->Tok.is(tok::comment)) { 2474 distributeComments(Comments, FormatTok); 2475 Comments.clear(); 2476 return; 2477 } 2478 2479 Comments.push_back(FormatTok); 2480 } while (!eof()); 2481 2482 distributeComments(Comments, nullptr); 2483 Comments.clear(); 2484 } 2485 2486 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2487 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2488 if (MustBreakBeforeNextToken) { 2489 Line->Tokens.back().Tok->MustBreakBefore = true; 2490 MustBreakBeforeNextToken = false; 2491 } 2492 } 2493 2494 } // end namespace format 2495 } // end namespace clang 2496