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 && isJSDeclOrStmt(Keywords, Next)) 857 return addUnwrappedLine(); 858 } 859 860 void UnwrappedLineParser::parseStructuralElement() { 861 assert(!FormatTok->is(tok::l_brace)); 862 if (Style.Language == FormatStyle::LK_TableGen && 863 FormatTok->is(tok::pp_include)) { 864 nextToken(); 865 if (FormatTok->is(tok::string_literal)) 866 nextToken(); 867 addUnwrappedLine(); 868 return; 869 } 870 switch (FormatTok->Tok.getKind()) { 871 case tok::at: 872 nextToken(); 873 if (FormatTok->Tok.is(tok::l_brace)) { 874 nextToken(); 875 parseBracedList(); 876 break; 877 } 878 switch (FormatTok->Tok.getObjCKeywordID()) { 879 case tok::objc_public: 880 case tok::objc_protected: 881 case tok::objc_package: 882 case tok::objc_private: 883 return parseAccessSpecifier(); 884 case tok::objc_interface: 885 case tok::objc_implementation: 886 return parseObjCInterfaceOrImplementation(); 887 case tok::objc_protocol: 888 return parseObjCProtocol(); 889 case tok::objc_end: 890 return; // Handled by the caller. 891 case tok::objc_optional: 892 case tok::objc_required: 893 nextToken(); 894 addUnwrappedLine(); 895 return; 896 case tok::objc_autoreleasepool: 897 nextToken(); 898 if (FormatTok->Tok.is(tok::l_brace)) { 899 if (Style.BraceWrapping.AfterObjCDeclaration) 900 addUnwrappedLine(); 901 parseBlock(/*MustBeDeclaration=*/false); 902 } 903 addUnwrappedLine(); 904 return; 905 case tok::objc_try: 906 // This branch isn't strictly necessary (the kw_try case below would 907 // do this too after the tok::at is parsed above). But be explicit. 908 parseTryCatch(); 909 return; 910 default: 911 break; 912 } 913 break; 914 case tok::kw_asm: 915 nextToken(); 916 if (FormatTok->is(tok::l_brace)) { 917 FormatTok->Type = TT_InlineASMBrace; 918 nextToken(); 919 while (FormatTok && FormatTok->isNot(tok::eof)) { 920 if (FormatTok->is(tok::r_brace)) { 921 FormatTok->Type = TT_InlineASMBrace; 922 nextToken(); 923 addUnwrappedLine(); 924 break; 925 } 926 FormatTok->Finalized = true; 927 nextToken(); 928 } 929 } 930 break; 931 case tok::kw_namespace: 932 parseNamespace(); 933 return; 934 case tok::kw_inline: 935 nextToken(); 936 if (FormatTok->Tok.is(tok::kw_namespace)) { 937 parseNamespace(); 938 return; 939 } 940 break; 941 case tok::kw_public: 942 case tok::kw_protected: 943 case tok::kw_private: 944 if (Style.Language == FormatStyle::LK_Java || 945 Style.Language == FormatStyle::LK_JavaScript) 946 nextToken(); 947 else 948 parseAccessSpecifier(); 949 return; 950 case tok::kw_if: 951 parseIfThenElse(); 952 return; 953 case tok::kw_for: 954 case tok::kw_while: 955 parseForOrWhileLoop(); 956 return; 957 case tok::kw_do: 958 parseDoWhile(); 959 return; 960 case tok::kw_switch: 961 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 962 // 'switch: string' field declaration. 963 break; 964 parseSwitch(); 965 return; 966 case tok::kw_default: 967 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 968 // 'default: string' field declaration. 969 break; 970 nextToken(); 971 parseLabel(); 972 return; 973 case tok::kw_case: 974 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 975 // 'case: string' field declaration. 976 break; 977 parseCaseLabel(); 978 return; 979 case tok::kw_try: 980 case tok::kw___try: 981 parseTryCatch(); 982 return; 983 case tok::kw_extern: 984 nextToken(); 985 if (FormatTok->Tok.is(tok::string_literal)) { 986 nextToken(); 987 if (FormatTok->Tok.is(tok::l_brace)) { 988 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 989 addUnwrappedLine(); 990 return; 991 } 992 } 993 break; 994 case tok::kw_export: 995 if (Style.Language == FormatStyle::LK_JavaScript) { 996 parseJavaScriptEs6ImportExport(); 997 return; 998 } 999 break; 1000 case tok::identifier: 1001 if (FormatTok->is(TT_ForEachMacro)) { 1002 parseForOrWhileLoop(); 1003 return; 1004 } 1005 if (FormatTok->is(TT_MacroBlockBegin)) { 1006 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1007 /*MunchSemi=*/false); 1008 return; 1009 } 1010 if (FormatTok->is(Keywords.kw_import)) { 1011 if (Style.Language == FormatStyle::LK_JavaScript) { 1012 parseJavaScriptEs6ImportExport(); 1013 return; 1014 } 1015 if (Style.Language == FormatStyle::LK_Proto) { 1016 nextToken(); 1017 if (FormatTok->is(tok::kw_public)) 1018 nextToken(); 1019 if (!FormatTok->is(tok::string_literal)) 1020 return; 1021 nextToken(); 1022 if (FormatTok->is(tok::semi)) 1023 nextToken(); 1024 addUnwrappedLine(); 1025 return; 1026 } 1027 } 1028 if (Style.isCpp() && 1029 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1030 Keywords.kw_slots, Keywords.kw_qslots)) { 1031 nextToken(); 1032 if (FormatTok->is(tok::colon)) { 1033 nextToken(); 1034 addUnwrappedLine(); 1035 return; 1036 } 1037 } 1038 // In all other cases, parse the declaration. 1039 break; 1040 default: 1041 break; 1042 } 1043 do { 1044 const FormatToken *Previous = getPreviousToken(); 1045 switch (FormatTok->Tok.getKind()) { 1046 case tok::at: 1047 nextToken(); 1048 if (FormatTok->Tok.is(tok::l_brace)) { 1049 nextToken(); 1050 parseBracedList(); 1051 } 1052 break; 1053 case tok::kw_enum: 1054 // Ignore if this is part of "template <enum ...". 1055 if (Previous && Previous->is(tok::less)) { 1056 nextToken(); 1057 break; 1058 } 1059 1060 // parseEnum falls through and does not yet add an unwrapped line as an 1061 // enum definition can start a structural element. 1062 if (!parseEnum()) 1063 break; 1064 // This only applies for C++. 1065 if (!Style.isCpp()) { 1066 addUnwrappedLine(); 1067 return; 1068 } 1069 break; 1070 case tok::kw_typedef: 1071 nextToken(); 1072 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1073 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 1074 parseEnum(); 1075 break; 1076 case tok::kw_struct: 1077 case tok::kw_union: 1078 case tok::kw_class: 1079 // parseRecord falls through and does not yet add an unwrapped line as a 1080 // record declaration or definition can start a structural element. 1081 parseRecord(); 1082 // This does not apply for Java and JavaScript. 1083 if (Style.Language == FormatStyle::LK_Java || 1084 Style.Language == FormatStyle::LK_JavaScript) { 1085 if (FormatTok->is(tok::semi)) 1086 nextToken(); 1087 addUnwrappedLine(); 1088 return; 1089 } 1090 break; 1091 case tok::period: 1092 nextToken(); 1093 // In Java, classes have an implicit static member "class". 1094 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1095 FormatTok->is(tok::kw_class)) 1096 nextToken(); 1097 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1098 FormatTok->Tok.getIdentifierInfo()) 1099 // JavaScript only has pseudo keywords, all keywords are allowed to 1100 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1101 nextToken(); 1102 break; 1103 case tok::semi: 1104 nextToken(); 1105 addUnwrappedLine(); 1106 return; 1107 case tok::r_brace: 1108 addUnwrappedLine(); 1109 return; 1110 case tok::l_paren: 1111 parseParens(); 1112 break; 1113 case tok::kw_operator: 1114 nextToken(); 1115 if (FormatTok->isBinaryOperator()) 1116 nextToken(); 1117 break; 1118 case tok::caret: 1119 nextToken(); 1120 if (FormatTok->Tok.isAnyIdentifier() || 1121 FormatTok->isSimpleTypeSpecifier()) 1122 nextToken(); 1123 if (FormatTok->is(tok::l_paren)) 1124 parseParens(); 1125 if (FormatTok->is(tok::l_brace)) 1126 parseChildBlock(); 1127 break; 1128 case tok::l_brace: 1129 if (!tryToParseBracedList()) { 1130 // A block outside of parentheses must be the last part of a 1131 // structural element. 1132 // FIXME: Figure out cases where this is not true, and add projections 1133 // for them (the one we know is missing are lambdas). 1134 if (Style.BraceWrapping.AfterFunction) 1135 addUnwrappedLine(); 1136 FormatTok->Type = TT_FunctionLBrace; 1137 parseBlock(/*MustBeDeclaration=*/false); 1138 addUnwrappedLine(); 1139 return; 1140 } 1141 // Otherwise this was a braced init list, and the structural 1142 // element continues. 1143 break; 1144 case tok::kw_try: 1145 // We arrive here when parsing function-try blocks. 1146 parseTryCatch(); 1147 return; 1148 case tok::identifier: { 1149 if (FormatTok->is(TT_MacroBlockEnd)) { 1150 addUnwrappedLine(); 1151 return; 1152 } 1153 1154 // Function declarations (as opposed to function expressions) are parsed 1155 // on their own unwrapped line by continuing this loop. Function 1156 // expressions (functions that are not on their own line) must not create 1157 // a new unwrapped line, so they are special cased below. 1158 size_t TokenCount = Line->Tokens.size(); 1159 if (Style.Language == FormatStyle::LK_JavaScript && 1160 FormatTok->is(Keywords.kw_function) && 1161 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1162 Keywords.kw_async)))) { 1163 tryToParseJSFunction(); 1164 break; 1165 } 1166 if ((Style.Language == FormatStyle::LK_JavaScript || 1167 Style.Language == FormatStyle::LK_Java) && 1168 FormatTok->is(Keywords.kw_interface)) { 1169 if (Style.Language == FormatStyle::LK_JavaScript) { 1170 // In JavaScript/TypeScript, "interface" can be used as a standalone 1171 // identifier, e.g. in `var interface = 1;`. If "interface" is 1172 // followed by another identifier, it is very like to be an actual 1173 // interface declaration. 1174 unsigned StoredPosition = Tokens->getPosition(); 1175 FormatToken *Next = Tokens->getNextToken(); 1176 FormatTok = Tokens->setPosition(StoredPosition); 1177 if (Next && !mustBeJSIdent(Keywords, Next)) { 1178 nextToken(); 1179 break; 1180 } 1181 } 1182 parseRecord(); 1183 addUnwrappedLine(); 1184 return; 1185 } 1186 1187 // See if the following token should start a new unwrapped line. 1188 StringRef Text = FormatTok->TokenText; 1189 nextToken(); 1190 if (Line->Tokens.size() == 1 && 1191 // JS doesn't have macros, and within classes colons indicate fields, 1192 // not labels. 1193 Style.Language != FormatStyle::LK_JavaScript) { 1194 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1195 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1196 parseLabel(); 1197 return; 1198 } 1199 // Recognize function-like macro usages without trailing semicolon as 1200 // well as free-standing macros like Q_OBJECT. 1201 bool FunctionLike = FormatTok->is(tok::l_paren); 1202 if (FunctionLike) 1203 parseParens(); 1204 1205 bool FollowedByNewline = 1206 CommentsBeforeNextToken.empty() 1207 ? FormatTok->NewlinesBefore > 0 1208 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1209 1210 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1211 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1212 addUnwrappedLine(); 1213 return; 1214 } 1215 } 1216 break; 1217 } 1218 case tok::equal: 1219 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1220 // TT_JsFatArrow. The always start an expression or a child block if 1221 // followed by a curly. 1222 if (FormatTok->is(TT_JsFatArrow)) { 1223 nextToken(); 1224 if (FormatTok->is(tok::l_brace)) 1225 parseChildBlock(); 1226 break; 1227 } 1228 1229 nextToken(); 1230 if (FormatTok->Tok.is(tok::l_brace)) { 1231 nextToken(); 1232 parseBracedList(); 1233 } else if (Style.Language == FormatStyle::LK_Proto && 1234 FormatTok->Tok.is(tok::less)) { 1235 nextToken(); 1236 parseBracedList(/*ContinueOnSemicolons=*/false, 1237 /*ClosingBraceKind=*/tok::greater); 1238 } 1239 break; 1240 case tok::l_square: 1241 parseSquare(); 1242 break; 1243 case tok::kw_new: 1244 parseNew(); 1245 break; 1246 default: 1247 nextToken(); 1248 break; 1249 } 1250 } while (!eof()); 1251 } 1252 1253 bool UnwrappedLineParser::tryToParseLambda() { 1254 if (!Style.isCpp()) { 1255 nextToken(); 1256 return false; 1257 } 1258 const FormatToken* Previous = getPreviousToken(); 1259 if (Previous && 1260 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1261 tok::kw_delete) || 1262 Previous->closesScope() || Previous->isSimpleTypeSpecifier())) { 1263 nextToken(); 1264 return false; 1265 } 1266 assert(FormatTok->is(tok::l_square)); 1267 FormatToken &LSquare = *FormatTok; 1268 if (!tryToParseLambdaIntroducer()) 1269 return false; 1270 1271 while (FormatTok->isNot(tok::l_brace)) { 1272 if (FormatTok->isSimpleTypeSpecifier()) { 1273 nextToken(); 1274 continue; 1275 } 1276 switch (FormatTok->Tok.getKind()) { 1277 case tok::l_brace: 1278 break; 1279 case tok::l_paren: 1280 parseParens(); 1281 break; 1282 case tok::amp: 1283 case tok::star: 1284 case tok::kw_const: 1285 case tok::comma: 1286 case tok::less: 1287 case tok::greater: 1288 case tok::identifier: 1289 case tok::numeric_constant: 1290 case tok::coloncolon: 1291 case tok::kw_mutable: 1292 nextToken(); 1293 break; 1294 case tok::arrow: 1295 FormatTok->Type = TT_LambdaArrow; 1296 nextToken(); 1297 break; 1298 default: 1299 return true; 1300 } 1301 } 1302 LSquare.Type = TT_LambdaLSquare; 1303 parseChildBlock(); 1304 return true; 1305 } 1306 1307 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1308 nextToken(); 1309 if (FormatTok->is(tok::equal)) { 1310 nextToken(); 1311 if (FormatTok->is(tok::r_square)) { 1312 nextToken(); 1313 return true; 1314 } 1315 if (FormatTok->isNot(tok::comma)) 1316 return false; 1317 nextToken(); 1318 } else if (FormatTok->is(tok::amp)) { 1319 nextToken(); 1320 if (FormatTok->is(tok::r_square)) { 1321 nextToken(); 1322 return true; 1323 } 1324 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) { 1325 return false; 1326 } 1327 if (FormatTok->is(tok::comma)) 1328 nextToken(); 1329 } else if (FormatTok->is(tok::r_square)) { 1330 nextToken(); 1331 return true; 1332 } 1333 do { 1334 if (FormatTok->is(tok::amp)) 1335 nextToken(); 1336 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this)) 1337 return false; 1338 nextToken(); 1339 if (FormatTok->is(tok::ellipsis)) 1340 nextToken(); 1341 if (FormatTok->is(tok::comma)) { 1342 nextToken(); 1343 } else if (FormatTok->is(tok::r_square)) { 1344 nextToken(); 1345 return true; 1346 } else { 1347 return false; 1348 } 1349 } while (!eof()); 1350 return false; 1351 } 1352 1353 void UnwrappedLineParser::tryToParseJSFunction() { 1354 assert(FormatTok->is(Keywords.kw_function) || 1355 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1356 if (FormatTok->is(Keywords.kw_async)) 1357 nextToken(); 1358 // Consume "function". 1359 nextToken(); 1360 1361 // Consume * (generator function). Treat it like C++'s overloaded operators. 1362 if (FormatTok->is(tok::star)) { 1363 FormatTok->Type = TT_OverloadedOperator; 1364 nextToken(); 1365 } 1366 1367 // Consume function name. 1368 if (FormatTok->is(tok::identifier)) 1369 nextToken(); 1370 1371 if (FormatTok->isNot(tok::l_paren)) 1372 return; 1373 1374 // Parse formal parameter list. 1375 parseParens(); 1376 1377 if (FormatTok->is(tok::colon)) { 1378 // Parse a type definition. 1379 nextToken(); 1380 1381 // Eat the type declaration. For braced inline object types, balance braces, 1382 // otherwise just parse until finding an l_brace for the function body. 1383 if (FormatTok->is(tok::l_brace)) 1384 tryToParseBracedList(); 1385 else 1386 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1387 nextToken(); 1388 } 1389 1390 if (FormatTok->is(tok::semi)) 1391 return; 1392 1393 parseChildBlock(); 1394 } 1395 1396 bool UnwrappedLineParser::tryToParseBracedList() { 1397 if (FormatTok->BlockKind == BK_Unknown) 1398 calculateBraceTypes(); 1399 assert(FormatTok->BlockKind != BK_Unknown); 1400 if (FormatTok->BlockKind == BK_Block) 1401 return false; 1402 nextToken(); 1403 parseBracedList(); 1404 return true; 1405 } 1406 1407 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1408 tok::TokenKind ClosingBraceKind) { 1409 bool HasError = false; 1410 1411 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1412 // replace this by using parseAssigmentExpression() inside. 1413 do { 1414 if (Style.Language == FormatStyle::LK_JavaScript) { 1415 if (FormatTok->is(Keywords.kw_function) || 1416 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1417 tryToParseJSFunction(); 1418 continue; 1419 } 1420 if (FormatTok->is(TT_JsFatArrow)) { 1421 nextToken(); 1422 // Fat arrows can be followed by simple expressions or by child blocks 1423 // in curly braces. 1424 if (FormatTok->is(tok::l_brace)) { 1425 parseChildBlock(); 1426 continue; 1427 } 1428 } 1429 if (FormatTok->is(tok::l_brace)) { 1430 // Could be a method inside of a braced list `{a() { return 1; }}`. 1431 if (tryToParseBracedList()) 1432 continue; 1433 parseChildBlock(); 1434 } 1435 } 1436 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1437 nextToken(); 1438 return !HasError; 1439 } 1440 switch (FormatTok->Tok.getKind()) { 1441 case tok::caret: 1442 nextToken(); 1443 if (FormatTok->is(tok::l_brace)) { 1444 parseChildBlock(); 1445 } 1446 break; 1447 case tok::l_square: 1448 tryToParseLambda(); 1449 break; 1450 case tok::l_paren: 1451 parseParens(); 1452 // JavaScript can just have free standing methods and getters/setters in 1453 // object literals. Detect them by a "{" following ")". 1454 if (Style.Language == FormatStyle::LK_JavaScript) { 1455 if (FormatTok->is(tok::l_brace)) 1456 parseChildBlock(); 1457 break; 1458 } 1459 break; 1460 case tok::l_brace: 1461 // Assume there are no blocks inside a braced init list apart 1462 // from the ones we explicitly parse out (like lambdas). 1463 FormatTok->BlockKind = BK_BracedInit; 1464 nextToken(); 1465 parseBracedList(); 1466 break; 1467 case tok::less: 1468 if (Style.Language == FormatStyle::LK_Proto) { 1469 nextToken(); 1470 parseBracedList(/*ContinueOnSemicolons=*/false, 1471 /*ClosingBraceKind=*/tok::greater); 1472 } else { 1473 nextToken(); 1474 } 1475 break; 1476 case tok::semi: 1477 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1478 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1479 // used for error recovery if we have otherwise determined that this is 1480 // a braced list. 1481 if (Style.Language == FormatStyle::LK_JavaScript) { 1482 nextToken(); 1483 break; 1484 } 1485 HasError = true; 1486 if (!ContinueOnSemicolons) 1487 return !HasError; 1488 nextToken(); 1489 break; 1490 case tok::comma: 1491 nextToken(); 1492 break; 1493 default: 1494 nextToken(); 1495 break; 1496 } 1497 } while (!eof()); 1498 return false; 1499 } 1500 1501 void UnwrappedLineParser::parseParens() { 1502 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1503 nextToken(); 1504 do { 1505 switch (FormatTok->Tok.getKind()) { 1506 case tok::l_paren: 1507 parseParens(); 1508 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1509 parseChildBlock(); 1510 break; 1511 case tok::r_paren: 1512 nextToken(); 1513 return; 1514 case tok::r_brace: 1515 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1516 return; 1517 case tok::l_square: 1518 tryToParseLambda(); 1519 break; 1520 case tok::l_brace: 1521 if (!tryToParseBracedList()) 1522 parseChildBlock(); 1523 break; 1524 case tok::at: 1525 nextToken(); 1526 if (FormatTok->Tok.is(tok::l_brace)) { 1527 nextToken(); 1528 parseBracedList(); 1529 } 1530 break; 1531 case tok::kw_class: 1532 if (Style.Language == FormatStyle::LK_JavaScript) 1533 parseRecord(/*ParseAsExpr=*/true); 1534 else 1535 nextToken(); 1536 break; 1537 case tok::identifier: 1538 if (Style.Language == FormatStyle::LK_JavaScript && 1539 (FormatTok->is(Keywords.kw_function) || 1540 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1541 tryToParseJSFunction(); 1542 else 1543 nextToken(); 1544 break; 1545 default: 1546 nextToken(); 1547 break; 1548 } 1549 } while (!eof()); 1550 } 1551 1552 void UnwrappedLineParser::parseSquare() { 1553 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1554 if (tryToParseLambda()) 1555 return; 1556 do { 1557 switch (FormatTok->Tok.getKind()) { 1558 case tok::l_paren: 1559 parseParens(); 1560 break; 1561 case tok::r_square: 1562 nextToken(); 1563 return; 1564 case tok::r_brace: 1565 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1566 return; 1567 case tok::l_square: 1568 parseSquare(); 1569 break; 1570 case tok::l_brace: { 1571 if (!tryToParseBracedList()) 1572 parseChildBlock(); 1573 break; 1574 } 1575 case tok::at: 1576 nextToken(); 1577 if (FormatTok->Tok.is(tok::l_brace)) { 1578 nextToken(); 1579 parseBracedList(); 1580 } 1581 break; 1582 default: 1583 nextToken(); 1584 break; 1585 } 1586 } while (!eof()); 1587 } 1588 1589 void UnwrappedLineParser::parseIfThenElse() { 1590 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1591 nextToken(); 1592 if (FormatTok->Tok.is(tok::kw_constexpr)) 1593 nextToken(); 1594 if (FormatTok->Tok.is(tok::l_paren)) 1595 parseParens(); 1596 bool NeedsUnwrappedLine = false; 1597 if (FormatTok->Tok.is(tok::l_brace)) { 1598 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1599 parseBlock(/*MustBeDeclaration=*/false); 1600 if (Style.BraceWrapping.BeforeElse) 1601 addUnwrappedLine(); 1602 else 1603 NeedsUnwrappedLine = true; 1604 } else { 1605 addUnwrappedLine(); 1606 ++Line->Level; 1607 parseStructuralElement(); 1608 --Line->Level; 1609 } 1610 if (FormatTok->Tok.is(tok::kw_else)) { 1611 nextToken(); 1612 if (FormatTok->Tok.is(tok::l_brace)) { 1613 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1614 parseBlock(/*MustBeDeclaration=*/false); 1615 addUnwrappedLine(); 1616 } else if (FormatTok->Tok.is(tok::kw_if)) { 1617 parseIfThenElse(); 1618 } else { 1619 addUnwrappedLine(); 1620 ++Line->Level; 1621 parseStructuralElement(); 1622 if (FormatTok->is(tok::eof)) 1623 addUnwrappedLine(); 1624 --Line->Level; 1625 } 1626 } else if (NeedsUnwrappedLine) { 1627 addUnwrappedLine(); 1628 } 1629 } 1630 1631 void UnwrappedLineParser::parseTryCatch() { 1632 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1633 nextToken(); 1634 bool NeedsUnwrappedLine = false; 1635 if (FormatTok->is(tok::colon)) { 1636 // We are in a function try block, what comes is an initializer list. 1637 nextToken(); 1638 while (FormatTok->is(tok::identifier)) { 1639 nextToken(); 1640 if (FormatTok->is(tok::l_paren)) 1641 parseParens(); 1642 if (FormatTok->is(tok::comma)) 1643 nextToken(); 1644 } 1645 } 1646 // Parse try with resource. 1647 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1648 parseParens(); 1649 } 1650 if (FormatTok->is(tok::l_brace)) { 1651 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1652 parseBlock(/*MustBeDeclaration=*/false); 1653 if (Style.BraceWrapping.BeforeCatch) { 1654 addUnwrappedLine(); 1655 } else { 1656 NeedsUnwrappedLine = true; 1657 } 1658 } else if (!FormatTok->is(tok::kw_catch)) { 1659 // The C++ standard requires a compound-statement after a try. 1660 // If there's none, we try to assume there's a structuralElement 1661 // and try to continue. 1662 addUnwrappedLine(); 1663 ++Line->Level; 1664 parseStructuralElement(); 1665 --Line->Level; 1666 } 1667 while (1) { 1668 if (FormatTok->is(tok::at)) 1669 nextToken(); 1670 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1671 tok::kw___finally) || 1672 ((Style.Language == FormatStyle::LK_Java || 1673 Style.Language == FormatStyle::LK_JavaScript) && 1674 FormatTok->is(Keywords.kw_finally)) || 1675 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1676 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1677 break; 1678 nextToken(); 1679 while (FormatTok->isNot(tok::l_brace)) { 1680 if (FormatTok->is(tok::l_paren)) { 1681 parseParens(); 1682 continue; 1683 } 1684 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1685 return; 1686 nextToken(); 1687 } 1688 NeedsUnwrappedLine = false; 1689 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1690 parseBlock(/*MustBeDeclaration=*/false); 1691 if (Style.BraceWrapping.BeforeCatch) 1692 addUnwrappedLine(); 1693 else 1694 NeedsUnwrappedLine = true; 1695 } 1696 if (NeedsUnwrappedLine) 1697 addUnwrappedLine(); 1698 } 1699 1700 void UnwrappedLineParser::parseNamespace() { 1701 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1702 1703 const FormatToken &InitialToken = *FormatTok; 1704 nextToken(); 1705 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1706 nextToken(); 1707 if (FormatTok->Tok.is(tok::l_brace)) { 1708 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1709 addUnwrappedLine(); 1710 1711 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1712 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1713 DeclarationScopeStack.size() > 1); 1714 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1715 // Munch the semicolon after a namespace. This is more common than one would 1716 // think. Puttin the semicolon into its own line is very ugly. 1717 if (FormatTok->Tok.is(tok::semi)) 1718 nextToken(); 1719 addUnwrappedLine(); 1720 } 1721 // FIXME: Add error handling. 1722 } 1723 1724 void UnwrappedLineParser::parseNew() { 1725 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1726 nextToken(); 1727 if (Style.Language != FormatStyle::LK_Java) 1728 return; 1729 1730 // In Java, we can parse everything up to the parens, which aren't optional. 1731 do { 1732 // There should not be a ;, { or } before the new's open paren. 1733 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1734 return; 1735 1736 // Consume the parens. 1737 if (FormatTok->is(tok::l_paren)) { 1738 parseParens(); 1739 1740 // If there is a class body of an anonymous class, consume that as child. 1741 if (FormatTok->is(tok::l_brace)) 1742 parseChildBlock(); 1743 return; 1744 } 1745 nextToken(); 1746 } while (!eof()); 1747 } 1748 1749 void UnwrappedLineParser::parseForOrWhileLoop() { 1750 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1751 "'for', 'while' or foreach macro expected"); 1752 nextToken(); 1753 // JS' for await ( ... 1754 if (Style.Language == FormatStyle::LK_JavaScript && 1755 FormatTok->is(Keywords.kw_await)) 1756 nextToken(); 1757 if (FormatTok->Tok.is(tok::l_paren)) 1758 parseParens(); 1759 if (FormatTok->Tok.is(tok::l_brace)) { 1760 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1761 parseBlock(/*MustBeDeclaration=*/false); 1762 addUnwrappedLine(); 1763 } else { 1764 addUnwrappedLine(); 1765 ++Line->Level; 1766 parseStructuralElement(); 1767 --Line->Level; 1768 } 1769 } 1770 1771 void UnwrappedLineParser::parseDoWhile() { 1772 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1773 nextToken(); 1774 if (FormatTok->Tok.is(tok::l_brace)) { 1775 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1776 parseBlock(/*MustBeDeclaration=*/false); 1777 if (Style.BraceWrapping.IndentBraces) 1778 addUnwrappedLine(); 1779 } else { 1780 addUnwrappedLine(); 1781 ++Line->Level; 1782 parseStructuralElement(); 1783 --Line->Level; 1784 } 1785 1786 // FIXME: Add error handling. 1787 if (!FormatTok->Tok.is(tok::kw_while)) { 1788 addUnwrappedLine(); 1789 return; 1790 } 1791 1792 nextToken(); 1793 parseStructuralElement(); 1794 } 1795 1796 void UnwrappedLineParser::parseLabel() { 1797 nextToken(); 1798 unsigned OldLineLevel = Line->Level; 1799 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1800 --Line->Level; 1801 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1802 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1803 parseBlock(/*MustBeDeclaration=*/false); 1804 if (FormatTok->Tok.is(tok::kw_break)) { 1805 if (Style.BraceWrapping.AfterControlStatement) 1806 addUnwrappedLine(); 1807 parseStructuralElement(); 1808 } 1809 addUnwrappedLine(); 1810 } else { 1811 if (FormatTok->is(tok::semi)) 1812 nextToken(); 1813 addUnwrappedLine(); 1814 } 1815 Line->Level = OldLineLevel; 1816 if (FormatTok->isNot(tok::l_brace)) { 1817 parseStructuralElement(); 1818 addUnwrappedLine(); 1819 } 1820 } 1821 1822 void UnwrappedLineParser::parseCaseLabel() { 1823 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1824 // FIXME: fix handling of complex expressions here. 1825 do { 1826 nextToken(); 1827 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1828 parseLabel(); 1829 } 1830 1831 void UnwrappedLineParser::parseSwitch() { 1832 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1833 nextToken(); 1834 if (FormatTok->Tok.is(tok::l_paren)) 1835 parseParens(); 1836 if (FormatTok->Tok.is(tok::l_brace)) { 1837 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1838 parseBlock(/*MustBeDeclaration=*/false); 1839 addUnwrappedLine(); 1840 } else { 1841 addUnwrappedLine(); 1842 ++Line->Level; 1843 parseStructuralElement(); 1844 --Line->Level; 1845 } 1846 } 1847 1848 void UnwrappedLineParser::parseAccessSpecifier() { 1849 nextToken(); 1850 // Understand Qt's slots. 1851 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1852 nextToken(); 1853 // Otherwise, we don't know what it is, and we'd better keep the next token. 1854 if (FormatTok->Tok.is(tok::colon)) 1855 nextToken(); 1856 addUnwrappedLine(); 1857 } 1858 1859 bool UnwrappedLineParser::parseEnum() { 1860 // Won't be 'enum' for NS_ENUMs. 1861 if (FormatTok->Tok.is(tok::kw_enum)) 1862 nextToken(); 1863 1864 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1865 // declarations. An "enum" keyword followed by a colon would be a syntax 1866 // error and thus assume it is just an identifier. 1867 if (Style.Language == FormatStyle::LK_JavaScript && 1868 FormatTok->isOneOf(tok::colon, tok::question)) 1869 return false; 1870 1871 // Eat up enum class ... 1872 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1873 nextToken(); 1874 1875 while (FormatTok->Tok.getIdentifierInfo() || 1876 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1877 tok::greater, tok::comma, tok::question)) { 1878 nextToken(); 1879 // We can have macros or attributes in between 'enum' and the enum name. 1880 if (FormatTok->is(tok::l_paren)) 1881 parseParens(); 1882 if (FormatTok->is(tok::identifier)) { 1883 nextToken(); 1884 // If there are two identifiers in a row, this is likely an elaborate 1885 // return type. In Java, this can be "implements", etc. 1886 if (Style.isCpp() && FormatTok->is(tok::identifier)) 1887 return false; 1888 } 1889 } 1890 1891 // Just a declaration or something is wrong. 1892 if (FormatTok->isNot(tok::l_brace)) 1893 return true; 1894 FormatTok->BlockKind = BK_Block; 1895 1896 if (Style.Language == FormatStyle::LK_Java) { 1897 // Java enums are different. 1898 parseJavaEnumBody(); 1899 return true; 1900 } 1901 if (Style.Language == FormatStyle::LK_Proto) { 1902 parseBlock(/*MustBeDeclaration=*/true); 1903 return true; 1904 } 1905 1906 // Parse enum body. 1907 nextToken(); 1908 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1909 if (HasError) { 1910 if (FormatTok->is(tok::semi)) 1911 nextToken(); 1912 addUnwrappedLine(); 1913 } 1914 return true; 1915 1916 // There is no addUnwrappedLine() here so that we fall through to parsing a 1917 // structural element afterwards. Thus, in "enum A {} n, m;", 1918 // "} n, m;" will end up in one unwrapped line. 1919 } 1920 1921 void UnwrappedLineParser::parseJavaEnumBody() { 1922 // Determine whether the enum is simple, i.e. does not have a semicolon or 1923 // constants with class bodies. Simple enums can be formatted like braced 1924 // lists, contracted to a single line, etc. 1925 unsigned StoredPosition = Tokens->getPosition(); 1926 bool IsSimple = true; 1927 FormatToken *Tok = Tokens->getNextToken(); 1928 while (Tok) { 1929 if (Tok->is(tok::r_brace)) 1930 break; 1931 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1932 IsSimple = false; 1933 break; 1934 } 1935 // FIXME: This will also mark enums with braces in the arguments to enum 1936 // constants as "not simple". This is probably fine in practice, though. 1937 Tok = Tokens->getNextToken(); 1938 } 1939 FormatTok = Tokens->setPosition(StoredPosition); 1940 1941 if (IsSimple) { 1942 nextToken(); 1943 parseBracedList(); 1944 addUnwrappedLine(); 1945 return; 1946 } 1947 1948 // Parse the body of a more complex enum. 1949 // First add a line for everything up to the "{". 1950 nextToken(); 1951 addUnwrappedLine(); 1952 ++Line->Level; 1953 1954 // Parse the enum constants. 1955 while (FormatTok) { 1956 if (FormatTok->is(tok::l_brace)) { 1957 // Parse the constant's class body. 1958 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1959 /*MunchSemi=*/false); 1960 } else if (FormatTok->is(tok::l_paren)) { 1961 parseParens(); 1962 } else if (FormatTok->is(tok::comma)) { 1963 nextToken(); 1964 addUnwrappedLine(); 1965 } else if (FormatTok->is(tok::semi)) { 1966 nextToken(); 1967 addUnwrappedLine(); 1968 break; 1969 } else if (FormatTok->is(tok::r_brace)) { 1970 addUnwrappedLine(); 1971 break; 1972 } else { 1973 nextToken(); 1974 } 1975 } 1976 1977 // Parse the class body after the enum's ";" if any. 1978 parseLevel(/*HasOpeningBrace=*/true); 1979 nextToken(); 1980 --Line->Level; 1981 addUnwrappedLine(); 1982 } 1983 1984 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 1985 const FormatToken &InitialToken = *FormatTok; 1986 nextToken(); 1987 1988 // The actual identifier can be a nested name specifier, and in macros 1989 // it is often token-pasted. 1990 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 1991 tok::kw___attribute, tok::kw___declspec, 1992 tok::kw_alignas) || 1993 ((Style.Language == FormatStyle::LK_Java || 1994 Style.Language == FormatStyle::LK_JavaScript) && 1995 FormatTok->isOneOf(tok::period, tok::comma))) { 1996 if (Style.Language == FormatStyle::LK_JavaScript && 1997 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 1998 // JavaScript/TypeScript supports inline object types in 1999 // extends/implements positions: 2000 // class Foo implements {bar: number} { } 2001 nextToken(); 2002 if (FormatTok->is(tok::l_brace)) { 2003 tryToParseBracedList(); 2004 continue; 2005 } 2006 } 2007 bool IsNonMacroIdentifier = 2008 FormatTok->is(tok::identifier) && 2009 FormatTok->TokenText != FormatTok->TokenText.upper(); 2010 nextToken(); 2011 // We can have macros or attributes in between 'class' and the class name. 2012 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 2013 parseParens(); 2014 } 2015 2016 // Note that parsing away template declarations here leads to incorrectly 2017 // accepting function declarations as record declarations. 2018 // In general, we cannot solve this problem. Consider: 2019 // class A<int> B() {} 2020 // which can be a function definition or a class definition when B() is a 2021 // macro. If we find enough real-world cases where this is a problem, we 2022 // can parse for the 'template' keyword in the beginning of the statement, 2023 // and thus rule out the record production in case there is no template 2024 // (this would still leave us with an ambiguity between template function 2025 // and class declarations). 2026 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2027 while (!eof()) { 2028 if (FormatTok->is(tok::l_brace)) { 2029 calculateBraceTypes(/*ExpectClassBody=*/true); 2030 if (!tryToParseBracedList()) 2031 break; 2032 } 2033 if (FormatTok->Tok.is(tok::semi)) 2034 return; 2035 nextToken(); 2036 } 2037 } 2038 if (FormatTok->Tok.is(tok::l_brace)) { 2039 if (ParseAsExpr) { 2040 parseChildBlock(); 2041 } else { 2042 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2043 addUnwrappedLine(); 2044 2045 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2046 /*MunchSemi=*/false); 2047 } 2048 } 2049 // There is no addUnwrappedLine() here so that we fall through to parsing a 2050 // structural element afterwards. Thus, in "class A {} n, m;", 2051 // "} n, m;" will end up in one unwrapped line. 2052 } 2053 2054 void UnwrappedLineParser::parseObjCProtocolList() { 2055 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2056 do 2057 nextToken(); 2058 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2059 nextToken(); // Skip '>'. 2060 } 2061 2062 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2063 do { 2064 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2065 nextToken(); 2066 addUnwrappedLine(); 2067 break; 2068 } 2069 if (FormatTok->is(tok::l_brace)) { 2070 parseBlock(/*MustBeDeclaration=*/false); 2071 // In ObjC interfaces, nothing should be following the "}". 2072 addUnwrappedLine(); 2073 } else if (FormatTok->is(tok::r_brace)) { 2074 // Ignore stray "}". parseStructuralElement doesn't consume them. 2075 nextToken(); 2076 addUnwrappedLine(); 2077 } else { 2078 parseStructuralElement(); 2079 } 2080 } while (!eof()); 2081 } 2082 2083 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2084 nextToken(); 2085 nextToken(); // interface name 2086 2087 // @interface can be followed by either a base class, or a category. 2088 if (FormatTok->Tok.is(tok::colon)) { 2089 nextToken(); 2090 nextToken(); // base class name 2091 } else if (FormatTok->Tok.is(tok::l_paren)) 2092 // Skip category, if present. 2093 parseParens(); 2094 2095 if (FormatTok->Tok.is(tok::less)) 2096 parseObjCProtocolList(); 2097 2098 if (FormatTok->Tok.is(tok::l_brace)) { 2099 if (Style.BraceWrapping.AfterObjCDeclaration) 2100 addUnwrappedLine(); 2101 parseBlock(/*MustBeDeclaration=*/true); 2102 } 2103 2104 // With instance variables, this puts '}' on its own line. Without instance 2105 // variables, this ends the @interface line. 2106 addUnwrappedLine(); 2107 2108 parseObjCUntilAtEnd(); 2109 } 2110 2111 void UnwrappedLineParser::parseObjCProtocol() { 2112 nextToken(); 2113 nextToken(); // protocol name 2114 2115 if (FormatTok->Tok.is(tok::less)) 2116 parseObjCProtocolList(); 2117 2118 // Check for protocol declaration. 2119 if (FormatTok->Tok.is(tok::semi)) { 2120 nextToken(); 2121 return addUnwrappedLine(); 2122 } 2123 2124 addUnwrappedLine(); 2125 parseObjCUntilAtEnd(); 2126 } 2127 2128 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2129 bool IsImport = FormatTok->is(Keywords.kw_import); 2130 assert(IsImport || FormatTok->is(tok::kw_export)); 2131 nextToken(); 2132 2133 // Consume the "default" in "export default class/function". 2134 if (FormatTok->is(tok::kw_default)) 2135 nextToken(); 2136 2137 // Consume "async function", "function" and "default function", so that these 2138 // get parsed as free-standing JS functions, i.e. do not require a trailing 2139 // semicolon. 2140 if (FormatTok->is(Keywords.kw_async)) 2141 nextToken(); 2142 if (FormatTok->is(Keywords.kw_function)) { 2143 nextToken(); 2144 return; 2145 } 2146 2147 // For imports, `export *`, `export {...}`, consume the rest of the line up 2148 // to the terminating `;`. For everything else, just return and continue 2149 // parsing the structural element, i.e. the declaration or expression for 2150 // `export default`. 2151 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2152 !FormatTok->isStringLiteral()) 2153 return; 2154 2155 while (!eof()) { 2156 if (FormatTok->is(tok::semi)) 2157 return; 2158 if (Line->Tokens.size() == 0) { 2159 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2160 // import statement should terminate. 2161 return; 2162 } 2163 if (FormatTok->is(tok::l_brace)) { 2164 FormatTok->BlockKind = BK_Block; 2165 nextToken(); 2166 parseBracedList(); 2167 } else { 2168 nextToken(); 2169 } 2170 } 2171 } 2172 2173 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2174 StringRef Prefix = "") { 2175 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")" 2176 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2177 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2178 E = Line.Tokens.end(); 2179 I != E; ++I) { 2180 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2181 << "T=" << I->Tok->Type 2182 << ", OC=" << I->Tok->OriginalColumn << "] "; 2183 } 2184 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2185 E = Line.Tokens.end(); 2186 I != E; ++I) { 2187 const UnwrappedLineNode &Node = *I; 2188 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2189 I = Node.Children.begin(), 2190 E = Node.Children.end(); 2191 I != E; ++I) { 2192 printDebugInfo(*I, "\nChild: "); 2193 } 2194 } 2195 llvm::dbgs() << "\n"; 2196 } 2197 2198 void UnwrappedLineParser::addUnwrappedLine() { 2199 if (Line->Tokens.empty()) 2200 return; 2201 DEBUG({ 2202 if (CurrentLines == &Lines) 2203 printDebugInfo(*Line); 2204 }); 2205 CurrentLines->push_back(std::move(*Line)); 2206 Line->Tokens.clear(); 2207 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2208 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2209 CurrentLines->append( 2210 std::make_move_iterator(PreprocessorDirectives.begin()), 2211 std::make_move_iterator(PreprocessorDirectives.end())); 2212 PreprocessorDirectives.clear(); 2213 } 2214 } 2215 2216 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2217 2218 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2219 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2220 FormatTok.NewlinesBefore > 0; 2221 } 2222 2223 // Checks if \p FormatTok is a line comment that continues the line comment 2224 // section on \p Line. 2225 static bool continuesLineCommentSection(const FormatToken &FormatTok, 2226 const UnwrappedLine &Line, 2227 llvm::Regex &CommentPragmasRegex) { 2228 if (Line.Tokens.empty()) 2229 return false; 2230 2231 StringRef IndentContent = FormatTok.TokenText; 2232 if (FormatTok.TokenText.startswith("//") || 2233 FormatTok.TokenText.startswith("/*")) 2234 IndentContent = FormatTok.TokenText.substr(2); 2235 if (CommentPragmasRegex.match(IndentContent)) 2236 return false; 2237 2238 // If Line starts with a line comment, then FormatTok continues the comment 2239 // section if its original column is greater or equal to the original start 2240 // column of the line. 2241 // 2242 // Define the min column token of a line as follows: if a line ends in '{' or 2243 // contains a '{' followed by a line comment, then the min column token is 2244 // that '{'. Otherwise, the min column token of the line is the first token of 2245 // the line. 2246 // 2247 // If Line starts with a token other than a line comment, then FormatTok 2248 // continues the comment section if its original column is greater than the 2249 // original start column of the min column token of the line. 2250 // 2251 // For example, the second line comment continues the first in these cases: 2252 // 2253 // // first line 2254 // // second line 2255 // 2256 // and: 2257 // 2258 // // first line 2259 // // second line 2260 // 2261 // and: 2262 // 2263 // int i; // first line 2264 // // second line 2265 // 2266 // and: 2267 // 2268 // do { // first line 2269 // // second line 2270 // int i; 2271 // } while (true); 2272 // 2273 // and: 2274 // 2275 // enum { 2276 // a, // first line 2277 // // second line 2278 // b 2279 // }; 2280 // 2281 // The second line comment doesn't continue the first in these cases: 2282 // 2283 // // first line 2284 // // second line 2285 // 2286 // and: 2287 // 2288 // int i; // first line 2289 // // second line 2290 // 2291 // and: 2292 // 2293 // do { // first line 2294 // // second line 2295 // int i; 2296 // } while (true); 2297 // 2298 // and: 2299 // 2300 // enum { 2301 // a, // first line 2302 // // second line 2303 // }; 2304 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2305 2306 // Scan for '{//'. If found, use the column of '{' as a min column for line 2307 // comment section continuation. 2308 const FormatToken *PreviousToken = nullptr; 2309 for (const UnwrappedLineNode &Node : Line.Tokens) { 2310 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2311 isLineComment(*Node.Tok)) { 2312 MinColumnToken = PreviousToken; 2313 break; 2314 } 2315 PreviousToken = Node.Tok; 2316 2317 // Grab the last newline preceding a token in this unwrapped line. 2318 if (Node.Tok->NewlinesBefore > 0) { 2319 MinColumnToken = Node.Tok; 2320 } 2321 } 2322 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2323 MinColumnToken = PreviousToken; 2324 } 2325 2326 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2327 MinColumnToken); 2328 } 2329 2330 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2331 bool JustComments = Line->Tokens.empty(); 2332 for (SmallVectorImpl<FormatToken *>::const_iterator 2333 I = CommentsBeforeNextToken.begin(), 2334 E = CommentsBeforeNextToken.end(); 2335 I != E; ++I) { 2336 // Line comments that belong to the same line comment section are put on the 2337 // same line since later we might want to reflow content between them. 2338 // Additional fine-grained breaking of line comment sections is controlled 2339 // by the class BreakableLineCommentSection in case it is desirable to keep 2340 // several line comment sections in the same unwrapped line. 2341 // 2342 // FIXME: Consider putting separate line comment sections as children to the 2343 // unwrapped line instead. 2344 (*I)->ContinuesLineCommentSection = 2345 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2346 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2347 addUnwrappedLine(); 2348 pushToken(*I); 2349 } 2350 if (NewlineBeforeNext && JustComments) 2351 addUnwrappedLine(); 2352 CommentsBeforeNextToken.clear(); 2353 } 2354 2355 void UnwrappedLineParser::nextToken(int LevelDifference) { 2356 if (eof()) 2357 return; 2358 flushComments(isOnNewLine(*FormatTok)); 2359 pushToken(FormatTok); 2360 if (Style.Language != FormatStyle::LK_JavaScript) 2361 readToken(LevelDifference); 2362 else 2363 readTokenWithJavaScriptASI(); 2364 } 2365 2366 const FormatToken *UnwrappedLineParser::getPreviousToken() { 2367 // FIXME: This is a dirty way to access the previous token. Find a better 2368 // solution. 2369 if (!Line || Line->Tokens.empty()) 2370 return nullptr; 2371 return Line->Tokens.back().Tok; 2372 } 2373 2374 void UnwrappedLineParser::distributeComments( 2375 const SmallVectorImpl<FormatToken *> &Comments, 2376 const FormatToken *NextTok) { 2377 // Whether or not a line comment token continues a line is controlled by 2378 // the method continuesLineCommentSection, with the following caveat: 2379 // 2380 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2381 // that each comment line from the trail is aligned with the next token, if 2382 // the next token exists. If a trail exists, the beginning of the maximal 2383 // trail is marked as a start of a new comment section. 2384 // 2385 // For example in this code: 2386 // 2387 // int a; // line about a 2388 // // line 1 about b 2389 // // line 2 about b 2390 // int b; 2391 // 2392 // the two lines about b form a maximal trail, so there are two sections, the 2393 // first one consisting of the single comment "// line about a" and the 2394 // second one consisting of the next two comments. 2395 if (Comments.empty()) 2396 return; 2397 bool ShouldPushCommentsInCurrentLine = true; 2398 bool HasTrailAlignedWithNextToken = false; 2399 unsigned StartOfTrailAlignedWithNextToken = 0; 2400 if (NextTok) { 2401 // We are skipping the first element intentionally. 2402 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2403 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2404 HasTrailAlignedWithNextToken = true; 2405 StartOfTrailAlignedWithNextToken = i; 2406 } 2407 } 2408 } 2409 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2410 FormatToken *FormatTok = Comments[i]; 2411 if (HasTrailAlignedWithNextToken && 2412 i == StartOfTrailAlignedWithNextToken) { 2413 FormatTok->ContinuesLineCommentSection = false; 2414 } else { 2415 FormatTok->ContinuesLineCommentSection = 2416 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2417 } 2418 if (!FormatTok->ContinuesLineCommentSection && 2419 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2420 ShouldPushCommentsInCurrentLine = false; 2421 } 2422 if (ShouldPushCommentsInCurrentLine) { 2423 pushToken(FormatTok); 2424 } else { 2425 CommentsBeforeNextToken.push_back(FormatTok); 2426 } 2427 } 2428 } 2429 2430 void UnwrappedLineParser::readToken(int LevelDifference) { 2431 SmallVector<FormatToken *, 1> Comments; 2432 do { 2433 FormatTok = Tokens->getNextToken(); 2434 assert(FormatTok); 2435 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2436 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2437 distributeComments(Comments, FormatTok); 2438 Comments.clear(); 2439 // If there is an unfinished unwrapped line, we flush the preprocessor 2440 // directives only after that unwrapped line was finished later. 2441 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2442 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2443 assert((LevelDifference >= 0 || 2444 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 2445 "LevelDifference makes Line->Level negative"); 2446 Line->Level += LevelDifference; 2447 // Comments stored before the preprocessor directive need to be output 2448 // before the preprocessor directive, at the same level as the 2449 // preprocessor directive, as we consider them to apply to the directive. 2450 flushComments(isOnNewLine(*FormatTok)); 2451 parsePPDirective(); 2452 } 2453 while (FormatTok->Type == TT_ConflictStart || 2454 FormatTok->Type == TT_ConflictEnd || 2455 FormatTok->Type == TT_ConflictAlternative) { 2456 if (FormatTok->Type == TT_ConflictStart) { 2457 conditionalCompilationStart(/*Unreachable=*/false); 2458 } else if (FormatTok->Type == TT_ConflictAlternative) { 2459 conditionalCompilationAlternative(); 2460 } else if (FormatTok->Type == TT_ConflictEnd) { 2461 conditionalCompilationEnd(); 2462 } 2463 FormatTok = Tokens->getNextToken(); 2464 FormatTok->MustBreakBefore = true; 2465 } 2466 2467 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 2468 !Line->InPPDirective) { 2469 continue; 2470 } 2471 2472 if (!FormatTok->Tok.is(tok::comment)) { 2473 distributeComments(Comments, FormatTok); 2474 Comments.clear(); 2475 return; 2476 } 2477 2478 Comments.push_back(FormatTok); 2479 } while (!eof()); 2480 2481 distributeComments(Comments, nullptr); 2482 Comments.clear(); 2483 } 2484 2485 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2486 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2487 if (MustBreakBeforeNextToken) { 2488 Line->Tokens.back().Tok->MustBreakBefore = true; 2489 MustBreakBeforeNextToken = false; 2490 } 2491 } 2492 2493 } // end namespace format 2494 } // end namespace clang 2495