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 #include <algorithm> 22 23 #define DEBUG_TYPE "format-parser" 24 25 namespace clang { 26 namespace format { 27 28 class FormatTokenSource { 29 public: 30 virtual ~FormatTokenSource() {} 31 virtual FormatToken *getNextToken() = 0; 32 33 virtual unsigned getPosition() = 0; 34 virtual FormatToken *setPosition(unsigned Position) = 0; 35 }; 36 37 namespace { 38 39 class ScopedDeclarationState { 40 public: 41 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack, 42 bool MustBeDeclaration) 43 : Line(Line), Stack(Stack) { 44 Line.MustBeDeclaration = MustBeDeclaration; 45 Stack.push_back(MustBeDeclaration); 46 } 47 ~ScopedDeclarationState() { 48 Stack.pop_back(); 49 if (!Stack.empty()) 50 Line.MustBeDeclaration = Stack.back(); 51 else 52 Line.MustBeDeclaration = true; 53 } 54 55 private: 56 UnwrappedLine &Line; 57 std::vector<bool> &Stack; 58 }; 59 60 static bool isLineComment(const FormatToken &FormatTok) { 61 return FormatTok.is(tok::comment) && !FormatTok.TokenText.startswith("/*"); 62 } 63 64 // Checks if \p FormatTok is a line comment that continues the line comment 65 // \p Previous. The original column of \p MinColumnToken is used to determine 66 // whether \p FormatTok is indented enough to the right to continue \p Previous. 67 static bool continuesLineComment(const FormatToken &FormatTok, 68 const FormatToken *Previous, 69 const FormatToken *MinColumnToken) { 70 if (!Previous || !MinColumnToken) 71 return false; 72 unsigned MinContinueColumn = 73 MinColumnToken->OriginalColumn + (isLineComment(*MinColumnToken) ? 0 : 1); 74 return isLineComment(FormatTok) && FormatTok.NewlinesBefore == 1 && 75 isLineComment(*Previous) && 76 FormatTok.OriginalColumn >= MinContinueColumn; 77 } 78 79 class ScopedMacroState : public FormatTokenSource { 80 public: 81 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource, 82 FormatToken *&ResetToken) 83 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken), 84 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource), 85 Token(nullptr), PreviousToken(nullptr) { 86 TokenSource = this; 87 Line.Level = 0; 88 Line.InPPDirective = true; 89 } 90 91 ~ScopedMacroState() override { 92 TokenSource = PreviousTokenSource; 93 ResetToken = Token; 94 Line.InPPDirective = false; 95 Line.Level = PreviousLineLevel; 96 } 97 98 FormatToken *getNextToken() override { 99 // The \c UnwrappedLineParser guards against this by never calling 100 // \c getNextToken() after it has encountered the first eof token. 101 assert(!eof()); 102 PreviousToken = Token; 103 Token = PreviousTokenSource->getNextToken(); 104 if (eof()) 105 return getFakeEOF(); 106 return Token; 107 } 108 109 unsigned getPosition() override { return PreviousTokenSource->getPosition(); } 110 111 FormatToken *setPosition(unsigned Position) override { 112 PreviousToken = nullptr; 113 Token = PreviousTokenSource->setPosition(Position); 114 return Token; 115 } 116 117 private: 118 bool eof() { 119 return Token && Token->HasUnescapedNewline && 120 !continuesLineComment(*Token, PreviousToken, 121 /*MinColumnToken=*/PreviousToken); 122 } 123 124 FormatToken *getFakeEOF() { 125 static bool EOFInitialized = false; 126 static FormatToken FormatTok; 127 if (!EOFInitialized) { 128 FormatTok.Tok.startToken(); 129 FormatTok.Tok.setKind(tok::eof); 130 EOFInitialized = true; 131 } 132 return &FormatTok; 133 } 134 135 UnwrappedLine &Line; 136 FormatTokenSource *&TokenSource; 137 FormatToken *&ResetToken; 138 unsigned PreviousLineLevel; 139 FormatTokenSource *PreviousTokenSource; 140 141 FormatToken *Token; 142 FormatToken *PreviousToken; 143 }; 144 145 } // end anonymous namespace 146 147 class ScopedLineState { 148 public: 149 ScopedLineState(UnwrappedLineParser &Parser, 150 bool SwitchToPreprocessorLines = false) 151 : Parser(Parser), OriginalLines(Parser.CurrentLines) { 152 if (SwitchToPreprocessorLines) 153 Parser.CurrentLines = &Parser.PreprocessorDirectives; 154 else if (!Parser.Line->Tokens.empty()) 155 Parser.CurrentLines = &Parser.Line->Tokens.back().Children; 156 PreBlockLine = std::move(Parser.Line); 157 Parser.Line = llvm::make_unique<UnwrappedLine>(); 158 Parser.Line->Level = PreBlockLine->Level; 159 Parser.Line->InPPDirective = PreBlockLine->InPPDirective; 160 } 161 162 ~ScopedLineState() { 163 if (!Parser.Line->Tokens.empty()) { 164 Parser.addUnwrappedLine(); 165 } 166 assert(Parser.Line->Tokens.empty()); 167 Parser.Line = std::move(PreBlockLine); 168 if (Parser.CurrentLines == &Parser.PreprocessorDirectives) 169 Parser.MustBreakBeforeNextToken = true; 170 Parser.CurrentLines = OriginalLines; 171 } 172 173 private: 174 UnwrappedLineParser &Parser; 175 176 std::unique_ptr<UnwrappedLine> PreBlockLine; 177 SmallVectorImpl<UnwrappedLine> *OriginalLines; 178 }; 179 180 class CompoundStatementIndenter { 181 public: 182 CompoundStatementIndenter(UnwrappedLineParser *Parser, 183 const FormatStyle &Style, unsigned &LineLevel) 184 : LineLevel(LineLevel), OldLineLevel(LineLevel) { 185 if (Style.BraceWrapping.AfterControlStatement) 186 Parser->addUnwrappedLine(); 187 if (Style.BraceWrapping.IndentBraces) 188 ++LineLevel; 189 } 190 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; } 191 192 private: 193 unsigned &LineLevel; 194 unsigned OldLineLevel; 195 }; 196 197 namespace { 198 199 class IndexedTokenSource : public FormatTokenSource { 200 public: 201 IndexedTokenSource(ArrayRef<FormatToken *> Tokens) 202 : Tokens(Tokens), Position(-1) {} 203 204 FormatToken *getNextToken() override { 205 ++Position; 206 return Tokens[Position]; 207 } 208 209 unsigned getPosition() override { 210 assert(Position >= 0); 211 return Position; 212 } 213 214 FormatToken *setPosition(unsigned P) override { 215 Position = P; 216 return Tokens[Position]; 217 } 218 219 void reset() { Position = -1; } 220 221 private: 222 ArrayRef<FormatToken *> Tokens; 223 int Position; 224 }; 225 226 } // end anonymous namespace 227 228 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style, 229 const AdditionalKeywords &Keywords, 230 unsigned FirstStartColumn, 231 ArrayRef<FormatToken *> Tokens, 232 UnwrappedLineConsumer &Callback) 233 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false), 234 CurrentLines(&Lines), Style(Style), Keywords(Keywords), 235 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr), 236 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1), 237 IfNdefCondition(nullptr), FoundIncludeGuardStart(false), 238 IncludeGuardRejected(false), FirstStartColumn(FirstStartColumn) {} 239 240 void UnwrappedLineParser::reset() { 241 PPBranchLevel = -1; 242 IfNdefCondition = nullptr; 243 FoundIncludeGuardStart = false; 244 IncludeGuardRejected = false; 245 Line.reset(new UnwrappedLine); 246 CommentsBeforeNextToken.clear(); 247 FormatTok = nullptr; 248 MustBreakBeforeNextToken = false; 249 PreprocessorDirectives.clear(); 250 CurrentLines = &Lines; 251 DeclarationScopeStack.clear(); 252 PPStack.clear(); 253 Line->FirstStartColumn = FirstStartColumn; 254 } 255 256 void UnwrappedLineParser::parse() { 257 IndexedTokenSource TokenSource(AllTokens); 258 Line->FirstStartColumn = FirstStartColumn; 259 do { 260 DEBUG(llvm::dbgs() << "----\n"); 261 reset(); 262 Tokens = &TokenSource; 263 TokenSource.reset(); 264 265 readToken(); 266 parseFile(); 267 // Create line with eof token. 268 pushToken(FormatTok); 269 addUnwrappedLine(); 270 271 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(), 272 E = Lines.end(); 273 I != E; ++I) { 274 Callback.consumeUnwrappedLine(*I); 275 } 276 Callback.finishRun(); 277 Lines.clear(); 278 while (!PPLevelBranchIndex.empty() && 279 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) { 280 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1); 281 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1); 282 } 283 if (!PPLevelBranchIndex.empty()) { 284 ++PPLevelBranchIndex.back(); 285 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size()); 286 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back()); 287 } 288 } while (!PPLevelBranchIndex.empty()); 289 } 290 291 void UnwrappedLineParser::parseFile() { 292 // The top-level context in a file always has declarations, except for pre- 293 // processor directives and JavaScript files. 294 bool MustBeDeclaration = 295 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript; 296 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 297 MustBeDeclaration); 298 if (Style.Language == FormatStyle::LK_TextProto) 299 parseBracedList(); 300 else 301 parseLevel(/*HasOpeningBrace=*/false); 302 // Make sure to format the remaining tokens. 303 flushComments(true); 304 addUnwrappedLine(); 305 } 306 307 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) { 308 bool SwitchLabelEncountered = false; 309 do { 310 tok::TokenKind kind = FormatTok->Tok.getKind(); 311 if (FormatTok->Type == TT_MacroBlockBegin) { 312 kind = tok::l_brace; 313 } else if (FormatTok->Type == TT_MacroBlockEnd) { 314 kind = tok::r_brace; 315 } 316 317 switch (kind) { 318 case tok::comment: 319 nextToken(); 320 addUnwrappedLine(); 321 break; 322 case tok::l_brace: 323 // FIXME: Add parameter whether this can happen - if this happens, we must 324 // be in a non-declaration context. 325 if (!FormatTok->is(TT_MacroBlockBegin) && tryToParseBracedList()) 326 continue; 327 parseBlock(/*MustBeDeclaration=*/false); 328 addUnwrappedLine(); 329 break; 330 case tok::r_brace: 331 if (HasOpeningBrace) 332 return; 333 nextToken(); 334 addUnwrappedLine(); 335 break; 336 case tok::kw_default: 337 case tok::kw_case: 338 if (Style.Language == FormatStyle::LK_JavaScript && 339 Line->MustBeDeclaration) { 340 // A 'case: string' style field declaration. 341 parseStructuralElement(); 342 break; 343 } 344 if (!SwitchLabelEncountered && 345 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1))) 346 ++Line->Level; 347 SwitchLabelEncountered = true; 348 parseStructuralElement(); 349 break; 350 default: 351 parseStructuralElement(); 352 break; 353 } 354 } while (!eof()); 355 } 356 357 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) { 358 // We'll parse forward through the tokens until we hit 359 // a closing brace or eof - note that getNextToken() will 360 // parse macros, so this will magically work inside macro 361 // definitions, too. 362 unsigned StoredPosition = Tokens->getPosition(); 363 FormatToken *Tok = FormatTok; 364 const FormatToken *PrevTok = Tok->Previous; 365 // Keep a stack of positions of lbrace tokens. We will 366 // update information about whether an lbrace starts a 367 // braced init list or a different block during the loop. 368 SmallVector<FormatToken *, 8> LBraceStack; 369 assert(Tok->Tok.is(tok::l_brace)); 370 do { 371 // Get next non-comment token. 372 FormatToken *NextTok; 373 unsigned ReadTokens = 0; 374 do { 375 NextTok = Tokens->getNextToken(); 376 ++ReadTokens; 377 } while (NextTok->is(tok::comment)); 378 379 switch (Tok->Tok.getKind()) { 380 case tok::l_brace: 381 if (Style.Language == FormatStyle::LK_JavaScript && PrevTok) { 382 if (PrevTok->isOneOf(tok::colon, tok::less)) 383 // A ':' indicates this code is in a type, or a braced list 384 // following a label in an object literal ({a: {b: 1}}). 385 // A '<' could be an object used in a comparison, but that is nonsense 386 // code (can never return true), so more likely it is a generic type 387 // argument (`X<{a: string; b: number}>`). 388 // The code below could be confused by semicolons between the 389 // individual members in a type member list, which would normally 390 // trigger BK_Block. In both cases, this must be parsed as an inline 391 // braced init. 392 Tok->BlockKind = BK_BracedInit; 393 else if (PrevTok->is(tok::r_paren)) 394 // `) { }` can only occur in function or method declarations in JS. 395 Tok->BlockKind = BK_Block; 396 } else { 397 Tok->BlockKind = BK_Unknown; 398 } 399 LBraceStack.push_back(Tok); 400 break; 401 case tok::r_brace: 402 if (LBraceStack.empty()) 403 break; 404 if (LBraceStack.back()->BlockKind == BK_Unknown) { 405 bool ProbablyBracedList = false; 406 if (Style.Language == FormatStyle::LK_Proto) { 407 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square); 408 } else { 409 // Using OriginalColumn to distinguish between ObjC methods and 410 // binary operators is a bit hacky. 411 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) && 412 NextTok->OriginalColumn == 0; 413 414 // If there is a comma, semicolon or right paren after the closing 415 // brace, we assume this is a braced initializer list. Note that 416 // regardless how we mark inner braces here, we will overwrite the 417 // BlockKind later if we parse a braced list (where all blocks 418 // inside are by default braced lists), or when we explicitly detect 419 // blocks (for example while parsing lambdas). 420 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a 421 // braced list in JS. 422 ProbablyBracedList = 423 (Style.Language == FormatStyle::LK_JavaScript && 424 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in, 425 Keywords.kw_as)) || 426 (Style.isCpp() && NextTok->is(tok::l_paren)) || 427 NextTok->isOneOf(tok::comma, tok::period, tok::colon, 428 tok::r_paren, tok::r_square, tok::l_brace, 429 tok::l_square, tok::ellipsis) || 430 (NextTok->is(tok::identifier) && 431 !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace)) || 432 (NextTok->is(tok::semi) && 433 (!ExpectClassBody || LBraceStack.size() != 1)) || 434 (NextTok->isBinaryOperator() && !NextIsObjCMethod); 435 } 436 if (ProbablyBracedList) { 437 Tok->BlockKind = BK_BracedInit; 438 LBraceStack.back()->BlockKind = BK_BracedInit; 439 } else { 440 Tok->BlockKind = BK_Block; 441 LBraceStack.back()->BlockKind = BK_Block; 442 } 443 } 444 LBraceStack.pop_back(); 445 break; 446 case tok::at: 447 case tok::semi: 448 case tok::kw_if: 449 case tok::kw_while: 450 case tok::kw_for: 451 case tok::kw_switch: 452 case tok::kw_try: 453 case tok::kw___try: 454 if (!LBraceStack.empty() && LBraceStack.back()->BlockKind == BK_Unknown) 455 LBraceStack.back()->BlockKind = BK_Block; 456 break; 457 default: 458 break; 459 } 460 PrevTok = Tok; 461 Tok = NextTok; 462 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty()); 463 464 // Assume other blocks for all unclosed opening braces. 465 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) { 466 if (LBraceStack[i]->BlockKind == BK_Unknown) 467 LBraceStack[i]->BlockKind = BK_Block; 468 } 469 470 FormatTok = Tokens->setPosition(StoredPosition); 471 } 472 473 template <class T> 474 static inline void hash_combine(std::size_t &seed, const T &v) { 475 std::hash<T> hasher; 476 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2); 477 } 478 479 size_t UnwrappedLineParser::computePPHash() const { 480 size_t h = 0; 481 for (const auto &i : PPStack) { 482 hash_combine(h, size_t(i.Kind)); 483 hash_combine(h, i.Line); 484 } 485 return h; 486 } 487 488 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel, 489 bool MunchSemi) { 490 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) && 491 "'{' or macro block token expected"); 492 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin); 493 FormatTok->BlockKind = BK_Block; 494 495 size_t PPStartHash = computePPHash(); 496 497 unsigned InitialLevel = Line->Level; 498 nextToken(/*LevelDifference=*/AddLevel ? 1 : 0); 499 500 if (MacroBlock && FormatTok->is(tok::l_paren)) 501 parseParens(); 502 503 size_t NbPreprocessorDirectives = 504 CurrentLines == &Lines ? PreprocessorDirectives.size() : 0; 505 addUnwrappedLine(); 506 size_t OpeningLineIndex = 507 CurrentLines->empty() 508 ? (UnwrappedLine::kInvalidIndex) 509 : (CurrentLines->size() - 1 - NbPreprocessorDirectives); 510 511 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 512 MustBeDeclaration); 513 if (AddLevel) 514 ++Line->Level; 515 parseLevel(/*HasOpeningBrace=*/true); 516 517 if (eof()) 518 return; 519 520 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd) 521 : !FormatTok->is(tok::r_brace)) { 522 Line->Level = InitialLevel; 523 FormatTok->BlockKind = BK_Block; 524 return; 525 } 526 527 size_t PPEndHash = computePPHash(); 528 529 // Munch the closing brace. 530 nextToken(/*LevelDifference=*/AddLevel ? -1 : 0); 531 532 if (MacroBlock && FormatTok->is(tok::l_paren)) 533 parseParens(); 534 535 if (MunchSemi && FormatTok->Tok.is(tok::semi)) 536 nextToken(); 537 Line->Level = InitialLevel; 538 539 if (PPStartHash == PPEndHash) { 540 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex; 541 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) { 542 // Update the opening line to add the forward reference as well 543 (*CurrentLines)[OpeningLineIndex].MatchingOpeningBlockLineIndex = 544 CurrentLines->size() - 1; 545 } 546 } 547 } 548 549 static bool isGoogScope(const UnwrappedLine &Line) { 550 // FIXME: Closure-library specific stuff should not be hard-coded but be 551 // configurable. 552 if (Line.Tokens.size() < 4) 553 return false; 554 auto I = Line.Tokens.begin(); 555 if (I->Tok->TokenText != "goog") 556 return false; 557 ++I; 558 if (I->Tok->isNot(tok::period)) 559 return false; 560 ++I; 561 if (I->Tok->TokenText != "scope") 562 return false; 563 ++I; 564 return I->Tok->is(tok::l_paren); 565 } 566 567 static bool isIIFE(const UnwrappedLine &Line, 568 const AdditionalKeywords &Keywords) { 569 // Look for the start of an immediately invoked anonymous function. 570 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression 571 // This is commonly done in JavaScript to create a new, anonymous scope. 572 // Example: (function() { ... })() 573 if (Line.Tokens.size() < 3) 574 return false; 575 auto I = Line.Tokens.begin(); 576 if (I->Tok->isNot(tok::l_paren)) 577 return false; 578 ++I; 579 if (I->Tok->isNot(Keywords.kw_function)) 580 return false; 581 ++I; 582 return I->Tok->is(tok::l_paren); 583 } 584 585 static bool ShouldBreakBeforeBrace(const FormatStyle &Style, 586 const FormatToken &InitialToken) { 587 if (InitialToken.is(tok::kw_namespace)) 588 return Style.BraceWrapping.AfterNamespace; 589 if (InitialToken.is(tok::kw_class)) 590 return Style.BraceWrapping.AfterClass; 591 if (InitialToken.is(tok::kw_union)) 592 return Style.BraceWrapping.AfterUnion; 593 if (InitialToken.is(tok::kw_struct)) 594 return Style.BraceWrapping.AfterStruct; 595 return false; 596 } 597 598 void UnwrappedLineParser::parseChildBlock() { 599 FormatTok->BlockKind = BK_Block; 600 nextToken(); 601 { 602 bool SkipIndent = (Style.Language == FormatStyle::LK_JavaScript && 603 (isGoogScope(*Line) || isIIFE(*Line, Keywords))); 604 ScopedLineState LineState(*this); 605 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack, 606 /*MustBeDeclaration=*/false); 607 Line->Level += SkipIndent ? 0 : 1; 608 parseLevel(/*HasOpeningBrace=*/true); 609 flushComments(isOnNewLine(*FormatTok)); 610 Line->Level -= SkipIndent ? 0 : 1; 611 } 612 nextToken(); 613 } 614 615 void UnwrappedLineParser::parsePPDirective() { 616 assert(FormatTok->Tok.is(tok::hash) && "'#' expected"); 617 ScopedMacroState MacroState(*Line, Tokens, FormatTok); 618 nextToken(); 619 620 if (!FormatTok->Tok.getIdentifierInfo()) { 621 parsePPUnknown(); 622 return; 623 } 624 625 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) { 626 case tok::pp_define: 627 parsePPDefine(); 628 return; 629 case tok::pp_if: 630 parsePPIf(/*IfDef=*/false); 631 break; 632 case tok::pp_ifdef: 633 case tok::pp_ifndef: 634 parsePPIf(/*IfDef=*/true); 635 break; 636 case tok::pp_else: 637 parsePPElse(); 638 break; 639 case tok::pp_elif: 640 parsePPElIf(); 641 break; 642 case tok::pp_endif: 643 parsePPEndIf(); 644 break; 645 default: 646 parsePPUnknown(); 647 break; 648 } 649 } 650 651 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) { 652 size_t Line = CurrentLines->size(); 653 if (CurrentLines == &PreprocessorDirectives) 654 Line += Lines.size(); 655 656 if (Unreachable || 657 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) 658 PPStack.push_back({PP_Unreachable, Line}); 659 else 660 PPStack.push_back({PP_Conditional, Line}); 661 } 662 663 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) { 664 ++PPBranchLevel; 665 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size()); 666 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) { 667 PPLevelBranchIndex.push_back(0); 668 PPLevelBranchCount.push_back(0); 669 } 670 PPChainBranchIndex.push(0); 671 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0; 672 conditionalCompilationCondition(Unreachable || Skip); 673 } 674 675 void UnwrappedLineParser::conditionalCompilationAlternative() { 676 if (!PPStack.empty()) 677 PPStack.pop_back(); 678 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 679 if (!PPChainBranchIndex.empty()) 680 ++PPChainBranchIndex.top(); 681 conditionalCompilationCondition( 682 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() && 683 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top()); 684 } 685 686 void UnwrappedLineParser::conditionalCompilationEnd() { 687 assert(PPBranchLevel < (int)PPLevelBranchIndex.size()); 688 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) { 689 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) { 690 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1; 691 } 692 } 693 // Guard against #endif's without #if. 694 if (PPBranchLevel > -1) 695 --PPBranchLevel; 696 if (!PPChainBranchIndex.empty()) 697 PPChainBranchIndex.pop(); 698 if (!PPStack.empty()) 699 PPStack.pop_back(); 700 } 701 702 void UnwrappedLineParser::parsePPIf(bool IfDef) { 703 bool IfNDef = FormatTok->is(tok::pp_ifndef); 704 nextToken(); 705 bool Unreachable = false; 706 if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0")) 707 Unreachable = true; 708 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG") 709 Unreachable = true; 710 conditionalCompilationStart(Unreachable); 711 FormatToken *IfCondition = FormatTok; 712 // If there's a #ifndef on the first line, and the only lines before it are 713 // comments, it could be an include guard. 714 bool MaybeIncludeGuard = IfNDef; 715 if (!IncludeGuardRejected && !FoundIncludeGuardStart && MaybeIncludeGuard) { 716 for (auto &Line : Lines) { 717 if (!Line.Tokens.front().Tok->is(tok::comment)) { 718 MaybeIncludeGuard = false; 719 IncludeGuardRejected = true; 720 break; 721 } 722 } 723 } 724 --PPBranchLevel; 725 parsePPUnknown(); 726 ++PPBranchLevel; 727 if (!IncludeGuardRejected && !FoundIncludeGuardStart && MaybeIncludeGuard) 728 IfNdefCondition = IfCondition; 729 } 730 731 void UnwrappedLineParser::parsePPElse() { 732 // If a potential include guard has an #else, it's not an include guard. 733 if (FoundIncludeGuardStart && PPBranchLevel == 0) 734 FoundIncludeGuardStart = false; 735 conditionalCompilationAlternative(); 736 if (PPBranchLevel > -1) 737 --PPBranchLevel; 738 parsePPUnknown(); 739 ++PPBranchLevel; 740 } 741 742 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); } 743 744 void UnwrappedLineParser::parsePPEndIf() { 745 conditionalCompilationEnd(); 746 parsePPUnknown(); 747 // If the #endif of a potential include guard is the last thing in the file, 748 // then we count it as a real include guard and subtract one from every 749 // preprocessor indent. 750 unsigned TokenPosition = Tokens->getPosition(); 751 FormatToken *PeekNext = AllTokens[TokenPosition]; 752 if (FoundIncludeGuardStart && PPBranchLevel == -1 && PeekNext->is(tok::eof) && 753 Style.IndentPPDirectives != FormatStyle::PPDIS_None) 754 for (auto &Line : Lines) 755 if (Line.InPPDirective && Line.Level > 0) 756 --Line.Level; 757 } 758 759 void UnwrappedLineParser::parsePPDefine() { 760 nextToken(); 761 762 if (FormatTok->Tok.getKind() != tok::identifier) { 763 parsePPUnknown(); 764 return; 765 } 766 if (IfNdefCondition && IfNdefCondition->TokenText == FormatTok->TokenText) { 767 FoundIncludeGuardStart = true; 768 for (auto &Line : Lines) { 769 if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) { 770 FoundIncludeGuardStart = false; 771 break; 772 } 773 } 774 } 775 IfNdefCondition = nullptr; 776 nextToken(); 777 if (FormatTok->Tok.getKind() == tok::l_paren && 778 FormatTok->WhitespaceRange.getBegin() == 779 FormatTok->WhitespaceRange.getEnd()) { 780 parseParens(); 781 } 782 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) 783 Line->Level += PPBranchLevel + 1; 784 addUnwrappedLine(); 785 ++Line->Level; 786 787 // Errors during a preprocessor directive can only affect the layout of the 788 // preprocessor directive, and thus we ignore them. An alternative approach 789 // would be to use the same approach we use on the file level (no 790 // re-indentation if there was a structural error) within the macro 791 // definition. 792 parseFile(); 793 } 794 795 void UnwrappedLineParser::parsePPUnknown() { 796 do { 797 nextToken(); 798 } while (!eof()); 799 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) 800 Line->Level += PPBranchLevel + 1; 801 addUnwrappedLine(); 802 IfNdefCondition = nullptr; 803 } 804 805 // Here we blacklist certain tokens that are not usually the first token in an 806 // unwrapped line. This is used in attempt to distinguish macro calls without 807 // trailing semicolons from other constructs split to several lines. 808 static bool tokenCanStartNewLine(const clang::Token &Tok) { 809 // Semicolon can be a null-statement, l_square can be a start of a macro or 810 // a C++11 attribute, but this doesn't seem to be common. 811 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) && 812 Tok.isNot(tok::l_square) && 813 // Tokens that can only be used as binary operators and a part of 814 // overloaded operator names. 815 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) && 816 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) && 817 Tok.isNot(tok::less) && Tok.isNot(tok::greater) && 818 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) && 819 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) && 820 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) && 821 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) && 822 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) && 823 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) && 824 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) && 825 Tok.isNot(tok::lesslessequal) && 826 // Colon is used in labels, base class lists, initializer lists, 827 // range-based for loops, ternary operator, but should never be the 828 // first token in an unwrapped line. 829 Tok.isNot(tok::colon) && 830 // 'noexcept' is a trailing annotation. 831 Tok.isNot(tok::kw_noexcept); 832 } 833 834 static bool mustBeJSIdent(const AdditionalKeywords &Keywords, 835 const FormatToken *FormatTok) { 836 // FIXME: This returns true for C/C++ keywords like 'struct'. 837 return FormatTok->is(tok::identifier) && 838 (FormatTok->Tok.getIdentifierInfo() == nullptr || 839 !FormatTok->isOneOf( 840 Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async, 841 Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally, 842 Keywords.kw_function, Keywords.kw_import, Keywords.kw_is, 843 Keywords.kw_let, Keywords.kw_var, tok::kw_const, 844 Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements, 845 Keywords.kw_instanceof, Keywords.kw_interface, Keywords.kw_throws, 846 Keywords.kw_from)); 847 } 848 849 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords, 850 const FormatToken *FormatTok) { 851 return FormatTok->Tok.isLiteral() || 852 FormatTok->isOneOf(tok::kw_true, tok::kw_false) || 853 mustBeJSIdent(Keywords, FormatTok); 854 } 855 856 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement 857 // when encountered after a value (see mustBeJSIdentOrValue). 858 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords, 859 const FormatToken *FormatTok) { 860 return FormatTok->isOneOf( 861 tok::kw_return, Keywords.kw_yield, 862 // conditionals 863 tok::kw_if, tok::kw_else, 864 // loops 865 tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break, 866 // switch/case 867 tok::kw_switch, tok::kw_case, 868 // exceptions 869 tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally, 870 // declaration 871 tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let, 872 Keywords.kw_async, Keywords.kw_function, 873 // import/export 874 Keywords.kw_import, tok::kw_export); 875 } 876 877 // readTokenWithJavaScriptASI reads the next token and terminates the current 878 // line if JavaScript Automatic Semicolon Insertion must 879 // happen between the current token and the next token. 880 // 881 // This method is conservative - it cannot cover all edge cases of JavaScript, 882 // but only aims to correctly handle certain well known cases. It *must not* 883 // return true in speculative cases. 884 void UnwrappedLineParser::readTokenWithJavaScriptASI() { 885 FormatToken *Previous = FormatTok; 886 readToken(); 887 FormatToken *Next = FormatTok; 888 889 bool IsOnSameLine = 890 CommentsBeforeNextToken.empty() 891 ? Next->NewlinesBefore == 0 892 : CommentsBeforeNextToken.front()->NewlinesBefore == 0; 893 if (IsOnSameLine) 894 return; 895 896 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous); 897 bool PreviousStartsTemplateExpr = 898 Previous->is(TT_TemplateString) && Previous->TokenText.endswith("${"); 899 if (PreviousMustBeValue || Previous->is(tok::r_paren)) { 900 // If the line contains an '@' sign, the previous token might be an 901 // annotation, which can precede another identifier/value. 902 bool HasAt = std::find_if(Line->Tokens.begin(), Line->Tokens.end(), 903 [](UnwrappedLineNode &LineNode) { 904 return LineNode.Tok->is(tok::at); 905 }) != Line->Tokens.end(); 906 if (HasAt) 907 return; 908 } 909 if (Next->is(tok::exclaim) && PreviousMustBeValue) 910 return addUnwrappedLine(); 911 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next); 912 bool NextEndsTemplateExpr = 913 Next->is(TT_TemplateString) && Next->TokenText.startswith("}"); 914 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr && 915 (PreviousMustBeValue || 916 Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus, 917 tok::minusminus))) 918 return addUnwrappedLine(); 919 if ((PreviousMustBeValue || Previous->is(tok::r_paren)) && 920 isJSDeclOrStmt(Keywords, Next)) 921 return addUnwrappedLine(); 922 } 923 924 void UnwrappedLineParser::parseStructuralElement() { 925 assert(!FormatTok->is(tok::l_brace)); 926 if (Style.Language == FormatStyle::LK_TableGen && 927 FormatTok->is(tok::pp_include)) { 928 nextToken(); 929 if (FormatTok->is(tok::string_literal)) 930 nextToken(); 931 addUnwrappedLine(); 932 return; 933 } 934 switch (FormatTok->Tok.getKind()) { 935 case tok::at: 936 nextToken(); 937 if (FormatTok->Tok.is(tok::l_brace)) { 938 nextToken(); 939 parseBracedList(); 940 break; 941 } 942 switch (FormatTok->Tok.getObjCKeywordID()) { 943 case tok::objc_public: 944 case tok::objc_protected: 945 case tok::objc_package: 946 case tok::objc_private: 947 return parseAccessSpecifier(); 948 case tok::objc_interface: 949 case tok::objc_implementation: 950 return parseObjCInterfaceOrImplementation(); 951 case tok::objc_protocol: 952 return parseObjCProtocol(); 953 case tok::objc_end: 954 return; // Handled by the caller. 955 case tok::objc_optional: 956 case tok::objc_required: 957 nextToken(); 958 addUnwrappedLine(); 959 return; 960 case tok::objc_autoreleasepool: 961 nextToken(); 962 if (FormatTok->Tok.is(tok::l_brace)) { 963 if (Style.BraceWrapping.AfterObjCDeclaration) 964 addUnwrappedLine(); 965 parseBlock(/*MustBeDeclaration=*/false); 966 } 967 addUnwrappedLine(); 968 return; 969 case tok::objc_try: 970 // This branch isn't strictly necessary (the kw_try case below would 971 // do this too after the tok::at is parsed above). But be explicit. 972 parseTryCatch(); 973 return; 974 default: 975 break; 976 } 977 break; 978 case tok::kw_asm: 979 nextToken(); 980 if (FormatTok->is(tok::l_brace)) { 981 FormatTok->Type = TT_InlineASMBrace; 982 nextToken(); 983 while (FormatTok && FormatTok->isNot(tok::eof)) { 984 if (FormatTok->is(tok::r_brace)) { 985 FormatTok->Type = TT_InlineASMBrace; 986 nextToken(); 987 addUnwrappedLine(); 988 break; 989 } 990 FormatTok->Finalized = true; 991 nextToken(); 992 } 993 } 994 break; 995 case tok::kw_namespace: 996 parseNamespace(); 997 return; 998 case tok::kw_inline: 999 nextToken(); 1000 if (FormatTok->Tok.is(tok::kw_namespace)) { 1001 parseNamespace(); 1002 return; 1003 } 1004 break; 1005 case tok::kw_public: 1006 case tok::kw_protected: 1007 case tok::kw_private: 1008 if (Style.Language == FormatStyle::LK_Java || 1009 Style.Language == FormatStyle::LK_JavaScript) 1010 nextToken(); 1011 else 1012 parseAccessSpecifier(); 1013 return; 1014 case tok::kw_if: 1015 parseIfThenElse(); 1016 return; 1017 case tok::kw_for: 1018 case tok::kw_while: 1019 parseForOrWhileLoop(); 1020 return; 1021 case tok::kw_do: 1022 parseDoWhile(); 1023 return; 1024 case tok::kw_switch: 1025 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1026 // 'switch: string' field declaration. 1027 break; 1028 parseSwitch(); 1029 return; 1030 case tok::kw_default: 1031 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1032 // 'default: string' field declaration. 1033 break; 1034 nextToken(); 1035 parseLabel(); 1036 return; 1037 case tok::kw_case: 1038 if (Style.Language == FormatStyle::LK_JavaScript && Line->MustBeDeclaration) 1039 // 'case: string' field declaration. 1040 break; 1041 parseCaseLabel(); 1042 return; 1043 case tok::kw_try: 1044 case tok::kw___try: 1045 parseTryCatch(); 1046 return; 1047 case tok::kw_extern: 1048 nextToken(); 1049 if (FormatTok->Tok.is(tok::string_literal)) { 1050 nextToken(); 1051 if (FormatTok->Tok.is(tok::l_brace)) { 1052 if (Style.BraceWrapping.AfterExternBlock) { 1053 addUnwrappedLine(); 1054 parseBlock(/*MustBeDeclaration=*/true); 1055 } else { 1056 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false); 1057 } 1058 addUnwrappedLine(); 1059 return; 1060 } 1061 } 1062 break; 1063 case tok::kw_export: 1064 if (Style.Language == FormatStyle::LK_JavaScript) { 1065 parseJavaScriptEs6ImportExport(); 1066 return; 1067 } 1068 break; 1069 case tok::identifier: 1070 if (FormatTok->is(TT_ForEachMacro)) { 1071 parseForOrWhileLoop(); 1072 return; 1073 } 1074 if (FormatTok->is(TT_MacroBlockBegin)) { 1075 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true, 1076 /*MunchSemi=*/false); 1077 return; 1078 } 1079 if (FormatTok->is(Keywords.kw_import)) { 1080 if (Style.Language == FormatStyle::LK_JavaScript) { 1081 parseJavaScriptEs6ImportExport(); 1082 return; 1083 } 1084 if (Style.Language == FormatStyle::LK_Proto) { 1085 nextToken(); 1086 if (FormatTok->is(tok::kw_public)) 1087 nextToken(); 1088 if (!FormatTok->is(tok::string_literal)) 1089 return; 1090 nextToken(); 1091 if (FormatTok->is(tok::semi)) 1092 nextToken(); 1093 addUnwrappedLine(); 1094 return; 1095 } 1096 } 1097 if (Style.isCpp() && 1098 FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals, 1099 Keywords.kw_slots, Keywords.kw_qslots)) { 1100 nextToken(); 1101 if (FormatTok->is(tok::colon)) { 1102 nextToken(); 1103 addUnwrappedLine(); 1104 return; 1105 } 1106 } 1107 // In all other cases, parse the declaration. 1108 break; 1109 default: 1110 break; 1111 } 1112 do { 1113 const FormatToken *Previous = FormatTok->Previous; 1114 switch (FormatTok->Tok.getKind()) { 1115 case tok::at: 1116 nextToken(); 1117 if (FormatTok->Tok.is(tok::l_brace)) { 1118 nextToken(); 1119 parseBracedList(); 1120 } 1121 break; 1122 case tok::kw_enum: 1123 // Ignore if this is part of "template <enum ...". 1124 if (Previous && Previous->is(tok::less)) { 1125 nextToken(); 1126 break; 1127 } 1128 1129 // parseEnum falls through and does not yet add an unwrapped line as an 1130 // enum definition can start a structural element. 1131 if (!parseEnum()) 1132 break; 1133 // This only applies for C++. 1134 if (!Style.isCpp()) { 1135 addUnwrappedLine(); 1136 return; 1137 } 1138 break; 1139 case tok::kw_typedef: 1140 nextToken(); 1141 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS, 1142 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS)) 1143 parseEnum(); 1144 break; 1145 case tok::kw_struct: 1146 case tok::kw_union: 1147 case tok::kw_class: 1148 // parseRecord falls through and does not yet add an unwrapped line as a 1149 // record declaration or definition can start a structural element. 1150 parseRecord(); 1151 // This does not apply for Java and JavaScript. 1152 if (Style.Language == FormatStyle::LK_Java || 1153 Style.Language == FormatStyle::LK_JavaScript) { 1154 if (FormatTok->is(tok::semi)) 1155 nextToken(); 1156 addUnwrappedLine(); 1157 return; 1158 } 1159 break; 1160 case tok::period: 1161 nextToken(); 1162 // In Java, classes have an implicit static member "class". 1163 if (Style.Language == FormatStyle::LK_Java && FormatTok && 1164 FormatTok->is(tok::kw_class)) 1165 nextToken(); 1166 if (Style.Language == FormatStyle::LK_JavaScript && FormatTok && 1167 FormatTok->Tok.getIdentifierInfo()) 1168 // JavaScript only has pseudo keywords, all keywords are allowed to 1169 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6 1170 nextToken(); 1171 break; 1172 case tok::semi: 1173 nextToken(); 1174 addUnwrappedLine(); 1175 return; 1176 case tok::r_brace: 1177 addUnwrappedLine(); 1178 return; 1179 case tok::l_paren: 1180 parseParens(); 1181 break; 1182 case tok::kw_operator: 1183 nextToken(); 1184 if (FormatTok->isBinaryOperator()) 1185 nextToken(); 1186 break; 1187 case tok::caret: 1188 nextToken(); 1189 if (FormatTok->Tok.isAnyIdentifier() || 1190 FormatTok->isSimpleTypeSpecifier()) 1191 nextToken(); 1192 if (FormatTok->is(tok::l_paren)) 1193 parseParens(); 1194 if (FormatTok->is(tok::l_brace)) 1195 parseChildBlock(); 1196 break; 1197 case tok::l_brace: 1198 if (!tryToParseBracedList()) { 1199 // A block outside of parentheses must be the last part of a 1200 // structural element. 1201 // FIXME: Figure out cases where this is not true, and add projections 1202 // for them (the one we know is missing are lambdas). 1203 if (Style.BraceWrapping.AfterFunction) 1204 addUnwrappedLine(); 1205 FormatTok->Type = TT_FunctionLBrace; 1206 parseBlock(/*MustBeDeclaration=*/false); 1207 addUnwrappedLine(); 1208 return; 1209 } 1210 // Otherwise this was a braced init list, and the structural 1211 // element continues. 1212 break; 1213 case tok::kw_try: 1214 // We arrive here when parsing function-try blocks. 1215 parseTryCatch(); 1216 return; 1217 case tok::identifier: { 1218 if (FormatTok->is(TT_MacroBlockEnd)) { 1219 addUnwrappedLine(); 1220 return; 1221 } 1222 1223 // Function declarations (as opposed to function expressions) are parsed 1224 // on their own unwrapped line by continuing this loop. Function 1225 // expressions (functions that are not on their own line) must not create 1226 // a new unwrapped line, so they are special cased below. 1227 size_t TokenCount = Line->Tokens.size(); 1228 if (Style.Language == FormatStyle::LK_JavaScript && 1229 FormatTok->is(Keywords.kw_function) && 1230 (TokenCount > 1 || (TokenCount == 1 && !Line->Tokens.front().Tok->is( 1231 Keywords.kw_async)))) { 1232 tryToParseJSFunction(); 1233 break; 1234 } 1235 if ((Style.Language == FormatStyle::LK_JavaScript || 1236 Style.Language == FormatStyle::LK_Java) && 1237 FormatTok->is(Keywords.kw_interface)) { 1238 if (Style.Language == FormatStyle::LK_JavaScript) { 1239 // In JavaScript/TypeScript, "interface" can be used as a standalone 1240 // identifier, e.g. in `var interface = 1;`. If "interface" is 1241 // followed by another identifier, it is very like to be an actual 1242 // interface declaration. 1243 unsigned StoredPosition = Tokens->getPosition(); 1244 FormatToken *Next = Tokens->getNextToken(); 1245 FormatTok = Tokens->setPosition(StoredPosition); 1246 if (Next && !mustBeJSIdent(Keywords, Next)) { 1247 nextToken(); 1248 break; 1249 } 1250 } 1251 parseRecord(); 1252 addUnwrappedLine(); 1253 return; 1254 } 1255 1256 // See if the following token should start a new unwrapped line. 1257 StringRef Text = FormatTok->TokenText; 1258 nextToken(); 1259 if (Line->Tokens.size() == 1 && 1260 // JS doesn't have macros, and within classes colons indicate fields, 1261 // not labels. 1262 Style.Language != FormatStyle::LK_JavaScript) { 1263 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) { 1264 Line->Tokens.begin()->Tok->MustBreakBefore = true; 1265 parseLabel(); 1266 return; 1267 } 1268 // Recognize function-like macro usages without trailing semicolon as 1269 // well as free-standing macros like Q_OBJECT. 1270 bool FunctionLike = FormatTok->is(tok::l_paren); 1271 if (FunctionLike) 1272 parseParens(); 1273 1274 bool FollowedByNewline = 1275 CommentsBeforeNextToken.empty() 1276 ? FormatTok->NewlinesBefore > 0 1277 : CommentsBeforeNextToken.front()->NewlinesBefore > 0; 1278 1279 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) && 1280 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) { 1281 addUnwrappedLine(); 1282 return; 1283 } 1284 } 1285 break; 1286 } 1287 case tok::equal: 1288 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType 1289 // TT_JsFatArrow. The always start an expression or a child block if 1290 // followed by a curly. 1291 if (FormatTok->is(TT_JsFatArrow)) { 1292 nextToken(); 1293 if (FormatTok->is(tok::l_brace)) 1294 parseChildBlock(); 1295 break; 1296 } 1297 1298 nextToken(); 1299 if (FormatTok->Tok.is(tok::l_brace)) { 1300 nextToken(); 1301 parseBracedList(); 1302 } else if (Style.Language == FormatStyle::LK_Proto && 1303 FormatTok->Tok.is(tok::less)) { 1304 nextToken(); 1305 parseBracedList(/*ContinueOnSemicolons=*/false, 1306 /*ClosingBraceKind=*/tok::greater); 1307 } 1308 break; 1309 case tok::l_square: 1310 parseSquare(); 1311 break; 1312 case tok::kw_new: 1313 parseNew(); 1314 break; 1315 default: 1316 nextToken(); 1317 break; 1318 } 1319 } while (!eof()); 1320 } 1321 1322 bool UnwrappedLineParser::tryToParseLambda() { 1323 if (!Style.isCpp()) { 1324 nextToken(); 1325 return false; 1326 } 1327 assert(FormatTok->is(tok::l_square)); 1328 FormatToken &LSquare = *FormatTok; 1329 if (!tryToParseLambdaIntroducer()) 1330 return false; 1331 1332 while (FormatTok->isNot(tok::l_brace)) { 1333 if (FormatTok->isSimpleTypeSpecifier()) { 1334 nextToken(); 1335 continue; 1336 } 1337 switch (FormatTok->Tok.getKind()) { 1338 case tok::l_brace: 1339 break; 1340 case tok::l_paren: 1341 parseParens(); 1342 break; 1343 case tok::amp: 1344 case tok::star: 1345 case tok::kw_const: 1346 case tok::comma: 1347 case tok::less: 1348 case tok::greater: 1349 case tok::identifier: 1350 case tok::numeric_constant: 1351 case tok::coloncolon: 1352 case tok::kw_mutable: 1353 nextToken(); 1354 break; 1355 case tok::arrow: 1356 FormatTok->Type = TT_LambdaArrow; 1357 nextToken(); 1358 break; 1359 default: 1360 return true; 1361 } 1362 } 1363 LSquare.Type = TT_LambdaLSquare; 1364 parseChildBlock(); 1365 return true; 1366 } 1367 1368 bool UnwrappedLineParser::tryToParseLambdaIntroducer() { 1369 const FormatToken *Previous = FormatTok->Previous; 1370 if (Previous && 1371 (Previous->isOneOf(tok::identifier, tok::kw_operator, tok::kw_new, 1372 tok::kw_delete) || 1373 FormatTok->isCppStructuredBinding(Style) || Previous->closesScope() || 1374 Previous->isSimpleTypeSpecifier())) { 1375 nextToken(); 1376 return false; 1377 } 1378 nextToken(); 1379 parseSquare(/*LambdaIntroducer=*/true); 1380 return true; 1381 } 1382 1383 void UnwrappedLineParser::tryToParseJSFunction() { 1384 assert(FormatTok->is(Keywords.kw_function) || 1385 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)); 1386 if (FormatTok->is(Keywords.kw_async)) 1387 nextToken(); 1388 // Consume "function". 1389 nextToken(); 1390 1391 // Consume * (generator function). Treat it like C++'s overloaded operators. 1392 if (FormatTok->is(tok::star)) { 1393 FormatTok->Type = TT_OverloadedOperator; 1394 nextToken(); 1395 } 1396 1397 // Consume function name. 1398 if (FormatTok->is(tok::identifier)) 1399 nextToken(); 1400 1401 if (FormatTok->isNot(tok::l_paren)) 1402 return; 1403 1404 // Parse formal parameter list. 1405 parseParens(); 1406 1407 if (FormatTok->is(tok::colon)) { 1408 // Parse a type definition. 1409 nextToken(); 1410 1411 // Eat the type declaration. For braced inline object types, balance braces, 1412 // otherwise just parse until finding an l_brace for the function body. 1413 if (FormatTok->is(tok::l_brace)) 1414 tryToParseBracedList(); 1415 else 1416 while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof()) 1417 nextToken(); 1418 } 1419 1420 if (FormatTok->is(tok::semi)) 1421 return; 1422 1423 parseChildBlock(); 1424 } 1425 1426 bool UnwrappedLineParser::tryToParseBracedList() { 1427 if (FormatTok->BlockKind == BK_Unknown) 1428 calculateBraceTypes(); 1429 assert(FormatTok->BlockKind != BK_Unknown); 1430 if (FormatTok->BlockKind == BK_Block) 1431 return false; 1432 nextToken(); 1433 parseBracedList(); 1434 return true; 1435 } 1436 1437 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons, 1438 tok::TokenKind ClosingBraceKind) { 1439 bool HasError = false; 1440 1441 // FIXME: Once we have an expression parser in the UnwrappedLineParser, 1442 // replace this by using parseAssigmentExpression() inside. 1443 do { 1444 if (Style.Language == FormatStyle::LK_JavaScript) { 1445 if (FormatTok->is(Keywords.kw_function) || 1446 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function)) { 1447 tryToParseJSFunction(); 1448 continue; 1449 } 1450 if (FormatTok->is(TT_JsFatArrow)) { 1451 nextToken(); 1452 // Fat arrows can be followed by simple expressions or by child blocks 1453 // in curly braces. 1454 if (FormatTok->is(tok::l_brace)) { 1455 parseChildBlock(); 1456 continue; 1457 } 1458 } 1459 if (FormatTok->is(tok::l_brace)) { 1460 // Could be a method inside of a braced list `{a() { return 1; }}`. 1461 if (tryToParseBracedList()) 1462 continue; 1463 parseChildBlock(); 1464 } 1465 } 1466 if (FormatTok->Tok.getKind() == ClosingBraceKind) { 1467 nextToken(); 1468 return !HasError; 1469 } 1470 switch (FormatTok->Tok.getKind()) { 1471 case tok::caret: 1472 nextToken(); 1473 if (FormatTok->is(tok::l_brace)) { 1474 parseChildBlock(); 1475 } 1476 break; 1477 case tok::l_square: 1478 tryToParseLambda(); 1479 break; 1480 case tok::l_paren: 1481 parseParens(); 1482 // JavaScript can just have free standing methods and getters/setters in 1483 // object literals. Detect them by a "{" following ")". 1484 if (Style.Language == FormatStyle::LK_JavaScript) { 1485 if (FormatTok->is(tok::l_brace)) 1486 parseChildBlock(); 1487 break; 1488 } 1489 break; 1490 case tok::l_brace: 1491 // Assume there are no blocks inside a braced init list apart 1492 // from the ones we explicitly parse out (like lambdas). 1493 FormatTok->BlockKind = BK_BracedInit; 1494 nextToken(); 1495 parseBracedList(); 1496 break; 1497 case tok::less: 1498 if (Style.Language == FormatStyle::LK_Proto) { 1499 nextToken(); 1500 parseBracedList(/*ContinueOnSemicolons=*/false, 1501 /*ClosingBraceKind=*/tok::greater); 1502 } else { 1503 nextToken(); 1504 } 1505 break; 1506 case tok::semi: 1507 // JavaScript (or more precisely TypeScript) can have semicolons in braced 1508 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be 1509 // used for error recovery if we have otherwise determined that this is 1510 // a braced list. 1511 if (Style.Language == FormatStyle::LK_JavaScript) { 1512 nextToken(); 1513 break; 1514 } 1515 HasError = true; 1516 if (!ContinueOnSemicolons) 1517 return !HasError; 1518 nextToken(); 1519 break; 1520 case tok::comma: 1521 nextToken(); 1522 break; 1523 default: 1524 nextToken(); 1525 break; 1526 } 1527 } while (!eof()); 1528 return false; 1529 } 1530 1531 void UnwrappedLineParser::parseParens() { 1532 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected."); 1533 nextToken(); 1534 do { 1535 switch (FormatTok->Tok.getKind()) { 1536 case tok::l_paren: 1537 parseParens(); 1538 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace)) 1539 parseChildBlock(); 1540 break; 1541 case tok::r_paren: 1542 nextToken(); 1543 return; 1544 case tok::r_brace: 1545 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1546 return; 1547 case tok::l_square: 1548 tryToParseLambda(); 1549 break; 1550 case tok::l_brace: 1551 if (!tryToParseBracedList()) 1552 parseChildBlock(); 1553 break; 1554 case tok::at: 1555 nextToken(); 1556 if (FormatTok->Tok.is(tok::l_brace)) { 1557 nextToken(); 1558 parseBracedList(); 1559 } 1560 break; 1561 case tok::kw_class: 1562 if (Style.Language == FormatStyle::LK_JavaScript) 1563 parseRecord(/*ParseAsExpr=*/true); 1564 else 1565 nextToken(); 1566 break; 1567 case tok::identifier: 1568 if (Style.Language == FormatStyle::LK_JavaScript && 1569 (FormatTok->is(Keywords.kw_function) || 1570 FormatTok->startsSequence(Keywords.kw_async, Keywords.kw_function))) 1571 tryToParseJSFunction(); 1572 else 1573 nextToken(); 1574 break; 1575 default: 1576 nextToken(); 1577 break; 1578 } 1579 } while (!eof()); 1580 } 1581 1582 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) { 1583 if (!LambdaIntroducer) { 1584 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected."); 1585 if (tryToParseLambda()) 1586 return; 1587 } 1588 do { 1589 switch (FormatTok->Tok.getKind()) { 1590 case tok::l_paren: 1591 parseParens(); 1592 break; 1593 case tok::r_square: 1594 nextToken(); 1595 return; 1596 case tok::r_brace: 1597 // A "}" inside parenthesis is an error if there wasn't a matching "{". 1598 return; 1599 case tok::l_square: 1600 parseSquare(); 1601 break; 1602 case tok::l_brace: { 1603 if (!tryToParseBracedList()) 1604 parseChildBlock(); 1605 break; 1606 } 1607 case tok::at: 1608 nextToken(); 1609 if (FormatTok->Tok.is(tok::l_brace)) { 1610 nextToken(); 1611 parseBracedList(); 1612 } 1613 break; 1614 default: 1615 nextToken(); 1616 break; 1617 } 1618 } while (!eof()); 1619 } 1620 1621 void UnwrappedLineParser::parseIfThenElse() { 1622 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected"); 1623 nextToken(); 1624 if (FormatTok->Tok.is(tok::kw_constexpr)) 1625 nextToken(); 1626 if (FormatTok->Tok.is(tok::l_paren)) 1627 parseParens(); 1628 bool NeedsUnwrappedLine = false; 1629 if (FormatTok->Tok.is(tok::l_brace)) { 1630 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1631 parseBlock(/*MustBeDeclaration=*/false); 1632 if (Style.BraceWrapping.BeforeElse) 1633 addUnwrappedLine(); 1634 else 1635 NeedsUnwrappedLine = true; 1636 } else { 1637 addUnwrappedLine(); 1638 ++Line->Level; 1639 parseStructuralElement(); 1640 --Line->Level; 1641 } 1642 if (FormatTok->Tok.is(tok::kw_else)) { 1643 nextToken(); 1644 if (FormatTok->Tok.is(tok::l_brace)) { 1645 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1646 parseBlock(/*MustBeDeclaration=*/false); 1647 addUnwrappedLine(); 1648 } else if (FormatTok->Tok.is(tok::kw_if)) { 1649 parseIfThenElse(); 1650 } else { 1651 addUnwrappedLine(); 1652 ++Line->Level; 1653 parseStructuralElement(); 1654 if (FormatTok->is(tok::eof)) 1655 addUnwrappedLine(); 1656 --Line->Level; 1657 } 1658 } else if (NeedsUnwrappedLine) { 1659 addUnwrappedLine(); 1660 } 1661 } 1662 1663 void UnwrappedLineParser::parseTryCatch() { 1664 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected"); 1665 nextToken(); 1666 bool NeedsUnwrappedLine = false; 1667 if (FormatTok->is(tok::colon)) { 1668 // We are in a function try block, what comes is an initializer list. 1669 nextToken(); 1670 while (FormatTok->is(tok::identifier)) { 1671 nextToken(); 1672 if (FormatTok->is(tok::l_paren)) 1673 parseParens(); 1674 if (FormatTok->is(tok::comma)) 1675 nextToken(); 1676 } 1677 } 1678 // Parse try with resource. 1679 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) { 1680 parseParens(); 1681 } 1682 if (FormatTok->is(tok::l_brace)) { 1683 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1684 parseBlock(/*MustBeDeclaration=*/false); 1685 if (Style.BraceWrapping.BeforeCatch) { 1686 addUnwrappedLine(); 1687 } else { 1688 NeedsUnwrappedLine = true; 1689 } 1690 } else if (!FormatTok->is(tok::kw_catch)) { 1691 // The C++ standard requires a compound-statement after a try. 1692 // If there's none, we try to assume there's a structuralElement 1693 // and try to continue. 1694 addUnwrappedLine(); 1695 ++Line->Level; 1696 parseStructuralElement(); 1697 --Line->Level; 1698 } 1699 while (1) { 1700 if (FormatTok->is(tok::at)) 1701 nextToken(); 1702 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except, 1703 tok::kw___finally) || 1704 ((Style.Language == FormatStyle::LK_Java || 1705 Style.Language == FormatStyle::LK_JavaScript) && 1706 FormatTok->is(Keywords.kw_finally)) || 1707 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) || 1708 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally)))) 1709 break; 1710 nextToken(); 1711 while (FormatTok->isNot(tok::l_brace)) { 1712 if (FormatTok->is(tok::l_paren)) { 1713 parseParens(); 1714 continue; 1715 } 1716 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) 1717 return; 1718 nextToken(); 1719 } 1720 NeedsUnwrappedLine = false; 1721 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1722 parseBlock(/*MustBeDeclaration=*/false); 1723 if (Style.BraceWrapping.BeforeCatch) 1724 addUnwrappedLine(); 1725 else 1726 NeedsUnwrappedLine = true; 1727 } 1728 if (NeedsUnwrappedLine) 1729 addUnwrappedLine(); 1730 } 1731 1732 void UnwrappedLineParser::parseNamespace() { 1733 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected"); 1734 1735 const FormatToken &InitialToken = *FormatTok; 1736 nextToken(); 1737 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon)) 1738 nextToken(); 1739 if (FormatTok->Tok.is(tok::l_brace)) { 1740 if (ShouldBreakBeforeBrace(Style, InitialToken)) 1741 addUnwrappedLine(); 1742 1743 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All || 1744 (Style.NamespaceIndentation == FormatStyle::NI_Inner && 1745 DeclarationScopeStack.size() > 1); 1746 parseBlock(/*MustBeDeclaration=*/true, AddLevel); 1747 // Munch the semicolon after a namespace. This is more common than one would 1748 // think. Puttin the semicolon into its own line is very ugly. 1749 if (FormatTok->Tok.is(tok::semi)) 1750 nextToken(); 1751 addUnwrappedLine(); 1752 } 1753 // FIXME: Add error handling. 1754 } 1755 1756 void UnwrappedLineParser::parseNew() { 1757 assert(FormatTok->is(tok::kw_new) && "'new' expected"); 1758 nextToken(); 1759 if (Style.Language != FormatStyle::LK_Java) 1760 return; 1761 1762 // In Java, we can parse everything up to the parens, which aren't optional. 1763 do { 1764 // There should not be a ;, { or } before the new's open paren. 1765 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace)) 1766 return; 1767 1768 // Consume the parens. 1769 if (FormatTok->is(tok::l_paren)) { 1770 parseParens(); 1771 1772 // If there is a class body of an anonymous class, consume that as child. 1773 if (FormatTok->is(tok::l_brace)) 1774 parseChildBlock(); 1775 return; 1776 } 1777 nextToken(); 1778 } while (!eof()); 1779 } 1780 1781 void UnwrappedLineParser::parseForOrWhileLoop() { 1782 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) && 1783 "'for', 'while' or foreach macro expected"); 1784 nextToken(); 1785 // JS' for await ( ... 1786 if (Style.Language == FormatStyle::LK_JavaScript && 1787 FormatTok->is(Keywords.kw_await)) 1788 nextToken(); 1789 if (FormatTok->Tok.is(tok::l_paren)) 1790 parseParens(); 1791 if (FormatTok->Tok.is(tok::l_brace)) { 1792 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1793 parseBlock(/*MustBeDeclaration=*/false); 1794 addUnwrappedLine(); 1795 } else { 1796 addUnwrappedLine(); 1797 ++Line->Level; 1798 parseStructuralElement(); 1799 --Line->Level; 1800 } 1801 } 1802 1803 void UnwrappedLineParser::parseDoWhile() { 1804 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected"); 1805 nextToken(); 1806 if (FormatTok->Tok.is(tok::l_brace)) { 1807 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1808 parseBlock(/*MustBeDeclaration=*/false); 1809 if (Style.BraceWrapping.IndentBraces) 1810 addUnwrappedLine(); 1811 } else { 1812 addUnwrappedLine(); 1813 ++Line->Level; 1814 parseStructuralElement(); 1815 --Line->Level; 1816 } 1817 1818 // FIXME: Add error handling. 1819 if (!FormatTok->Tok.is(tok::kw_while)) { 1820 addUnwrappedLine(); 1821 return; 1822 } 1823 1824 nextToken(); 1825 parseStructuralElement(); 1826 } 1827 1828 void UnwrappedLineParser::parseLabel() { 1829 nextToken(); 1830 unsigned OldLineLevel = Line->Level; 1831 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0)) 1832 --Line->Level; 1833 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) { 1834 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1835 parseBlock(/*MustBeDeclaration=*/false); 1836 if (FormatTok->Tok.is(tok::kw_break)) { 1837 if (Style.BraceWrapping.AfterControlStatement) 1838 addUnwrappedLine(); 1839 parseStructuralElement(); 1840 } 1841 addUnwrappedLine(); 1842 } else { 1843 if (FormatTok->is(tok::semi)) 1844 nextToken(); 1845 addUnwrappedLine(); 1846 } 1847 Line->Level = OldLineLevel; 1848 if (FormatTok->isNot(tok::l_brace)) { 1849 parseStructuralElement(); 1850 addUnwrappedLine(); 1851 } 1852 } 1853 1854 void UnwrappedLineParser::parseCaseLabel() { 1855 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected"); 1856 // FIXME: fix handling of complex expressions here. 1857 do { 1858 nextToken(); 1859 } while (!eof() && !FormatTok->Tok.is(tok::colon)); 1860 parseLabel(); 1861 } 1862 1863 void UnwrappedLineParser::parseSwitch() { 1864 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected"); 1865 nextToken(); 1866 if (FormatTok->Tok.is(tok::l_paren)) 1867 parseParens(); 1868 if (FormatTok->Tok.is(tok::l_brace)) { 1869 CompoundStatementIndenter Indenter(this, Style, Line->Level); 1870 parseBlock(/*MustBeDeclaration=*/false); 1871 addUnwrappedLine(); 1872 } else { 1873 addUnwrappedLine(); 1874 ++Line->Level; 1875 parseStructuralElement(); 1876 --Line->Level; 1877 } 1878 } 1879 1880 void UnwrappedLineParser::parseAccessSpecifier() { 1881 nextToken(); 1882 // Understand Qt's slots. 1883 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots)) 1884 nextToken(); 1885 // Otherwise, we don't know what it is, and we'd better keep the next token. 1886 if (FormatTok->Tok.is(tok::colon)) 1887 nextToken(); 1888 addUnwrappedLine(); 1889 } 1890 1891 bool UnwrappedLineParser::parseEnum() { 1892 // Won't be 'enum' for NS_ENUMs. 1893 if (FormatTok->Tok.is(tok::kw_enum)) 1894 nextToken(); 1895 1896 // In TypeScript, "enum" can also be used as property name, e.g. in interface 1897 // declarations. An "enum" keyword followed by a colon would be a syntax 1898 // error and thus assume it is just an identifier. 1899 if (Style.Language == FormatStyle::LK_JavaScript && 1900 FormatTok->isOneOf(tok::colon, tok::question)) 1901 return false; 1902 1903 // Eat up enum class ... 1904 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct)) 1905 nextToken(); 1906 1907 while (FormatTok->Tok.getIdentifierInfo() || 1908 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less, 1909 tok::greater, tok::comma, tok::question)) { 1910 nextToken(); 1911 // We can have macros or attributes in between 'enum' and the enum name. 1912 if (FormatTok->is(tok::l_paren)) 1913 parseParens(); 1914 if (FormatTok->is(tok::identifier)) { 1915 nextToken(); 1916 // If there are two identifiers in a row, this is likely an elaborate 1917 // return type. In Java, this can be "implements", etc. 1918 if (Style.isCpp() && FormatTok->is(tok::identifier)) 1919 return false; 1920 } 1921 } 1922 1923 // Just a declaration or something is wrong. 1924 if (FormatTok->isNot(tok::l_brace)) 1925 return true; 1926 FormatTok->BlockKind = BK_Block; 1927 1928 if (Style.Language == FormatStyle::LK_Java) { 1929 // Java enums are different. 1930 parseJavaEnumBody(); 1931 return true; 1932 } 1933 if (Style.Language == FormatStyle::LK_Proto) { 1934 parseBlock(/*MustBeDeclaration=*/true); 1935 return true; 1936 } 1937 1938 // Parse enum body. 1939 nextToken(); 1940 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true); 1941 if (HasError) { 1942 if (FormatTok->is(tok::semi)) 1943 nextToken(); 1944 addUnwrappedLine(); 1945 } 1946 return true; 1947 1948 // There is no addUnwrappedLine() here so that we fall through to parsing a 1949 // structural element afterwards. Thus, in "enum A {} n, m;", 1950 // "} n, m;" will end up in one unwrapped line. 1951 } 1952 1953 void UnwrappedLineParser::parseJavaEnumBody() { 1954 // Determine whether the enum is simple, i.e. does not have a semicolon or 1955 // constants with class bodies. Simple enums can be formatted like braced 1956 // lists, contracted to a single line, etc. 1957 unsigned StoredPosition = Tokens->getPosition(); 1958 bool IsSimple = true; 1959 FormatToken *Tok = Tokens->getNextToken(); 1960 while (Tok) { 1961 if (Tok->is(tok::r_brace)) 1962 break; 1963 if (Tok->isOneOf(tok::l_brace, tok::semi)) { 1964 IsSimple = false; 1965 break; 1966 } 1967 // FIXME: This will also mark enums with braces in the arguments to enum 1968 // constants as "not simple". This is probably fine in practice, though. 1969 Tok = Tokens->getNextToken(); 1970 } 1971 FormatTok = Tokens->setPosition(StoredPosition); 1972 1973 if (IsSimple) { 1974 nextToken(); 1975 parseBracedList(); 1976 addUnwrappedLine(); 1977 return; 1978 } 1979 1980 // Parse the body of a more complex enum. 1981 // First add a line for everything up to the "{". 1982 nextToken(); 1983 addUnwrappedLine(); 1984 ++Line->Level; 1985 1986 // Parse the enum constants. 1987 while (FormatTok) { 1988 if (FormatTok->is(tok::l_brace)) { 1989 // Parse the constant's class body. 1990 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 1991 /*MunchSemi=*/false); 1992 } else if (FormatTok->is(tok::l_paren)) { 1993 parseParens(); 1994 } else if (FormatTok->is(tok::comma)) { 1995 nextToken(); 1996 addUnwrappedLine(); 1997 } else if (FormatTok->is(tok::semi)) { 1998 nextToken(); 1999 addUnwrappedLine(); 2000 break; 2001 } else if (FormatTok->is(tok::r_brace)) { 2002 addUnwrappedLine(); 2003 break; 2004 } else { 2005 nextToken(); 2006 } 2007 } 2008 2009 // Parse the class body after the enum's ";" if any. 2010 parseLevel(/*HasOpeningBrace=*/true); 2011 nextToken(); 2012 --Line->Level; 2013 addUnwrappedLine(); 2014 } 2015 2016 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) { 2017 const FormatToken &InitialToken = *FormatTok; 2018 nextToken(); 2019 2020 // The actual identifier can be a nested name specifier, and in macros 2021 // it is often token-pasted. 2022 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash, 2023 tok::kw___attribute, tok::kw___declspec, 2024 tok::kw_alignas) || 2025 ((Style.Language == FormatStyle::LK_Java || 2026 Style.Language == FormatStyle::LK_JavaScript) && 2027 FormatTok->isOneOf(tok::period, tok::comma))) { 2028 if (Style.Language == FormatStyle::LK_JavaScript && 2029 FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) { 2030 // JavaScript/TypeScript supports inline object types in 2031 // extends/implements positions: 2032 // class Foo implements {bar: number} { } 2033 nextToken(); 2034 if (FormatTok->is(tok::l_brace)) { 2035 tryToParseBracedList(); 2036 continue; 2037 } 2038 } 2039 bool IsNonMacroIdentifier = 2040 FormatTok->is(tok::identifier) && 2041 FormatTok->TokenText != FormatTok->TokenText.upper(); 2042 nextToken(); 2043 // We can have macros or attributes in between 'class' and the class name. 2044 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren)) 2045 parseParens(); 2046 } 2047 2048 // Note that parsing away template declarations here leads to incorrectly 2049 // accepting function declarations as record declarations. 2050 // In general, we cannot solve this problem. Consider: 2051 // class A<int> B() {} 2052 // which can be a function definition or a class definition when B() is a 2053 // macro. If we find enough real-world cases where this is a problem, we 2054 // can parse for the 'template' keyword in the beginning of the statement, 2055 // and thus rule out the record production in case there is no template 2056 // (this would still leave us with an ambiguity between template function 2057 // and class declarations). 2058 if (FormatTok->isOneOf(tok::colon, tok::less)) { 2059 while (!eof()) { 2060 if (FormatTok->is(tok::l_brace)) { 2061 calculateBraceTypes(/*ExpectClassBody=*/true); 2062 if (!tryToParseBracedList()) 2063 break; 2064 } 2065 if (FormatTok->Tok.is(tok::semi)) 2066 return; 2067 nextToken(); 2068 } 2069 } 2070 if (FormatTok->Tok.is(tok::l_brace)) { 2071 if (ParseAsExpr) { 2072 parseChildBlock(); 2073 } else { 2074 if (ShouldBreakBeforeBrace(Style, InitialToken)) 2075 addUnwrappedLine(); 2076 2077 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true, 2078 /*MunchSemi=*/false); 2079 } 2080 } 2081 // There is no addUnwrappedLine() here so that we fall through to parsing a 2082 // structural element afterwards. Thus, in "class A {} n, m;", 2083 // "} n, m;" will end up in one unwrapped line. 2084 } 2085 2086 void UnwrappedLineParser::parseObjCProtocolList() { 2087 assert(FormatTok->Tok.is(tok::less) && "'<' expected."); 2088 do 2089 nextToken(); 2090 while (!eof() && FormatTok->Tok.isNot(tok::greater)); 2091 nextToken(); // Skip '>'. 2092 } 2093 2094 void UnwrappedLineParser::parseObjCUntilAtEnd() { 2095 do { 2096 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) { 2097 nextToken(); 2098 addUnwrappedLine(); 2099 break; 2100 } 2101 if (FormatTok->is(tok::l_brace)) { 2102 parseBlock(/*MustBeDeclaration=*/false); 2103 // In ObjC interfaces, nothing should be following the "}". 2104 addUnwrappedLine(); 2105 } else if (FormatTok->is(tok::r_brace)) { 2106 // Ignore stray "}". parseStructuralElement doesn't consume them. 2107 nextToken(); 2108 addUnwrappedLine(); 2109 } else { 2110 parseStructuralElement(); 2111 } 2112 } while (!eof()); 2113 } 2114 2115 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() { 2116 nextToken(); 2117 nextToken(); // interface name 2118 2119 // @interface can be followed by either a base class, or a category. 2120 if (FormatTok->Tok.is(tok::colon)) { 2121 nextToken(); 2122 nextToken(); // base class name 2123 } else if (FormatTok->Tok.is(tok::l_paren)) 2124 // Skip category, if present. 2125 parseParens(); 2126 2127 if (FormatTok->Tok.is(tok::less)) 2128 parseObjCProtocolList(); 2129 2130 if (FormatTok->Tok.is(tok::l_brace)) { 2131 if (Style.BraceWrapping.AfterObjCDeclaration) 2132 addUnwrappedLine(); 2133 parseBlock(/*MustBeDeclaration=*/true); 2134 } 2135 2136 // With instance variables, this puts '}' on its own line. Without instance 2137 // variables, this ends the @interface line. 2138 addUnwrappedLine(); 2139 2140 parseObjCUntilAtEnd(); 2141 } 2142 2143 void UnwrappedLineParser::parseObjCProtocol() { 2144 nextToken(); 2145 nextToken(); // protocol name 2146 2147 if (FormatTok->Tok.is(tok::less)) 2148 parseObjCProtocolList(); 2149 2150 // Check for protocol declaration. 2151 if (FormatTok->Tok.is(tok::semi)) { 2152 nextToken(); 2153 return addUnwrappedLine(); 2154 } 2155 2156 addUnwrappedLine(); 2157 parseObjCUntilAtEnd(); 2158 } 2159 2160 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() { 2161 bool IsImport = FormatTok->is(Keywords.kw_import); 2162 assert(IsImport || FormatTok->is(tok::kw_export)); 2163 nextToken(); 2164 2165 // Consume the "default" in "export default class/function". 2166 if (FormatTok->is(tok::kw_default)) 2167 nextToken(); 2168 2169 // Consume "async function", "function" and "default function", so that these 2170 // get parsed as free-standing JS functions, i.e. do not require a trailing 2171 // semicolon. 2172 if (FormatTok->is(Keywords.kw_async)) 2173 nextToken(); 2174 if (FormatTok->is(Keywords.kw_function)) { 2175 nextToken(); 2176 return; 2177 } 2178 2179 // For imports, `export *`, `export {...}`, consume the rest of the line up 2180 // to the terminating `;`. For everything else, just return and continue 2181 // parsing the structural element, i.e. the declaration or expression for 2182 // `export default`. 2183 if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) && 2184 !FormatTok->isStringLiteral()) 2185 return; 2186 2187 while (!eof()) { 2188 if (FormatTok->is(tok::semi)) 2189 return; 2190 if (Line->Tokens.empty()) { 2191 // Common issue: Automatic Semicolon Insertion wrapped the line, so the 2192 // import statement should terminate. 2193 return; 2194 } 2195 if (FormatTok->is(tok::l_brace)) { 2196 FormatTok->BlockKind = BK_Block; 2197 nextToken(); 2198 parseBracedList(); 2199 } else { 2200 nextToken(); 2201 } 2202 } 2203 } 2204 2205 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line, 2206 StringRef Prefix = "") { 2207 llvm::dbgs() << Prefix << "Line(" << Line.Level 2208 << ", FSC=" << Line.FirstStartColumn << ")" 2209 << (Line.InPPDirective ? " MACRO" : "") << ": "; 2210 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2211 E = Line.Tokens.end(); 2212 I != E; ++I) { 2213 llvm::dbgs() << I->Tok->Tok.getName() << "[" 2214 << "T=" << I->Tok->Type << ", OC=" << I->Tok->OriginalColumn 2215 << "] "; 2216 } 2217 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(), 2218 E = Line.Tokens.end(); 2219 I != E; ++I) { 2220 const UnwrappedLineNode &Node = *I; 2221 for (SmallVectorImpl<UnwrappedLine>::const_iterator 2222 I = Node.Children.begin(), 2223 E = Node.Children.end(); 2224 I != E; ++I) { 2225 printDebugInfo(*I, "\nChild: "); 2226 } 2227 } 2228 llvm::dbgs() << "\n"; 2229 } 2230 2231 void UnwrappedLineParser::addUnwrappedLine() { 2232 if (Line->Tokens.empty()) 2233 return; 2234 DEBUG({ 2235 if (CurrentLines == &Lines) 2236 printDebugInfo(*Line); 2237 }); 2238 CurrentLines->push_back(std::move(*Line)); 2239 Line->Tokens.clear(); 2240 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex; 2241 Line->FirstStartColumn = 0; 2242 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) { 2243 CurrentLines->append( 2244 std::make_move_iterator(PreprocessorDirectives.begin()), 2245 std::make_move_iterator(PreprocessorDirectives.end())); 2246 PreprocessorDirectives.clear(); 2247 } 2248 // Disconnect the current token from the last token on the previous line. 2249 FormatTok->Previous = nullptr; 2250 } 2251 2252 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); } 2253 2254 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) { 2255 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) && 2256 FormatTok.NewlinesBefore > 0; 2257 } 2258 2259 // Checks if \p FormatTok is a line comment that continues the line comment 2260 // section on \p Line. 2261 static bool continuesLineCommentSection(const FormatToken &FormatTok, 2262 const UnwrappedLine &Line, 2263 llvm::Regex &CommentPragmasRegex) { 2264 if (Line.Tokens.empty()) 2265 return false; 2266 2267 StringRef IndentContent = FormatTok.TokenText; 2268 if (FormatTok.TokenText.startswith("//") || 2269 FormatTok.TokenText.startswith("/*")) 2270 IndentContent = FormatTok.TokenText.substr(2); 2271 if (CommentPragmasRegex.match(IndentContent)) 2272 return false; 2273 2274 // If Line starts with a line comment, then FormatTok continues the comment 2275 // section if its original column is greater or equal to the original start 2276 // column of the line. 2277 // 2278 // Define the min column token of a line as follows: if a line ends in '{' or 2279 // contains a '{' followed by a line comment, then the min column token is 2280 // that '{'. Otherwise, the min column token of the line is the first token of 2281 // the line. 2282 // 2283 // If Line starts with a token other than a line comment, then FormatTok 2284 // continues the comment section if its original column is greater than the 2285 // original start column of the min column token of the line. 2286 // 2287 // For example, the second line comment continues the first in these cases: 2288 // 2289 // // first line 2290 // // second line 2291 // 2292 // and: 2293 // 2294 // // first line 2295 // // second line 2296 // 2297 // and: 2298 // 2299 // int i; // first line 2300 // // second line 2301 // 2302 // and: 2303 // 2304 // do { // first line 2305 // // second line 2306 // int i; 2307 // } while (true); 2308 // 2309 // and: 2310 // 2311 // enum { 2312 // a, // first line 2313 // // second line 2314 // b 2315 // }; 2316 // 2317 // The second line comment doesn't continue the first in these cases: 2318 // 2319 // // first line 2320 // // second line 2321 // 2322 // and: 2323 // 2324 // int i; // first line 2325 // // second line 2326 // 2327 // and: 2328 // 2329 // do { // first line 2330 // // second line 2331 // int i; 2332 // } while (true); 2333 // 2334 // and: 2335 // 2336 // enum { 2337 // a, // first line 2338 // // second line 2339 // }; 2340 const FormatToken *MinColumnToken = Line.Tokens.front().Tok; 2341 2342 // Scan for '{//'. If found, use the column of '{' as a min column for line 2343 // comment section continuation. 2344 const FormatToken *PreviousToken = nullptr; 2345 for (const UnwrappedLineNode &Node : Line.Tokens) { 2346 if (PreviousToken && PreviousToken->is(tok::l_brace) && 2347 isLineComment(*Node.Tok)) { 2348 MinColumnToken = PreviousToken; 2349 break; 2350 } 2351 PreviousToken = Node.Tok; 2352 2353 // Grab the last newline preceding a token in this unwrapped line. 2354 if (Node.Tok->NewlinesBefore > 0) { 2355 MinColumnToken = Node.Tok; 2356 } 2357 } 2358 if (PreviousToken && PreviousToken->is(tok::l_brace)) { 2359 MinColumnToken = PreviousToken; 2360 } 2361 2362 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok, 2363 MinColumnToken); 2364 } 2365 2366 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) { 2367 bool JustComments = Line->Tokens.empty(); 2368 for (SmallVectorImpl<FormatToken *>::const_iterator 2369 I = CommentsBeforeNextToken.begin(), 2370 E = CommentsBeforeNextToken.end(); 2371 I != E; ++I) { 2372 // Line comments that belong to the same line comment section are put on the 2373 // same line since later we might want to reflow content between them. 2374 // Additional fine-grained breaking of line comment sections is controlled 2375 // by the class BreakableLineCommentSection in case it is desirable to keep 2376 // several line comment sections in the same unwrapped line. 2377 // 2378 // FIXME: Consider putting separate line comment sections as children to the 2379 // unwrapped line instead. 2380 (*I)->ContinuesLineCommentSection = 2381 continuesLineCommentSection(**I, *Line, CommentPragmasRegex); 2382 if (isOnNewLine(**I) && JustComments && !(*I)->ContinuesLineCommentSection) 2383 addUnwrappedLine(); 2384 pushToken(*I); 2385 } 2386 if (NewlineBeforeNext && JustComments) 2387 addUnwrappedLine(); 2388 CommentsBeforeNextToken.clear(); 2389 } 2390 2391 void UnwrappedLineParser::nextToken(int LevelDifference) { 2392 if (eof()) 2393 return; 2394 flushComments(isOnNewLine(*FormatTok)); 2395 pushToken(FormatTok); 2396 FormatToken *Previous = FormatTok; 2397 if (Style.Language != FormatStyle::LK_JavaScript) 2398 readToken(LevelDifference); 2399 else 2400 readTokenWithJavaScriptASI(); 2401 FormatTok->Previous = Previous; 2402 } 2403 2404 void UnwrappedLineParser::distributeComments( 2405 const SmallVectorImpl<FormatToken *> &Comments, 2406 const FormatToken *NextTok) { 2407 // Whether or not a line comment token continues a line is controlled by 2408 // the method continuesLineCommentSection, with the following caveat: 2409 // 2410 // Define a trail of Comments to be a nonempty proper postfix of Comments such 2411 // that each comment line from the trail is aligned with the next token, if 2412 // the next token exists. If a trail exists, the beginning of the maximal 2413 // trail is marked as a start of a new comment section. 2414 // 2415 // For example in this code: 2416 // 2417 // int a; // line about a 2418 // // line 1 about b 2419 // // line 2 about b 2420 // int b; 2421 // 2422 // the two lines about b form a maximal trail, so there are two sections, the 2423 // first one consisting of the single comment "// line about a" and the 2424 // second one consisting of the next two comments. 2425 if (Comments.empty()) 2426 return; 2427 bool ShouldPushCommentsInCurrentLine = true; 2428 bool HasTrailAlignedWithNextToken = false; 2429 unsigned StartOfTrailAlignedWithNextToken = 0; 2430 if (NextTok) { 2431 // We are skipping the first element intentionally. 2432 for (unsigned i = Comments.size() - 1; i > 0; --i) { 2433 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) { 2434 HasTrailAlignedWithNextToken = true; 2435 StartOfTrailAlignedWithNextToken = i; 2436 } 2437 } 2438 } 2439 for (unsigned i = 0, e = Comments.size(); i < e; ++i) { 2440 FormatToken *FormatTok = Comments[i]; 2441 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) { 2442 FormatTok->ContinuesLineCommentSection = false; 2443 } else { 2444 FormatTok->ContinuesLineCommentSection = 2445 continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex); 2446 } 2447 if (!FormatTok->ContinuesLineCommentSection && 2448 (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) { 2449 ShouldPushCommentsInCurrentLine = false; 2450 } 2451 if (ShouldPushCommentsInCurrentLine) { 2452 pushToken(FormatTok); 2453 } else { 2454 CommentsBeforeNextToken.push_back(FormatTok); 2455 } 2456 } 2457 } 2458 2459 void UnwrappedLineParser::readToken(int LevelDifference) { 2460 SmallVector<FormatToken *, 1> Comments; 2461 do { 2462 FormatTok = Tokens->getNextToken(); 2463 assert(FormatTok); 2464 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) && 2465 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) { 2466 distributeComments(Comments, FormatTok); 2467 Comments.clear(); 2468 // If there is an unfinished unwrapped line, we flush the preprocessor 2469 // directives only after that unwrapped line was finished later. 2470 bool SwitchToPreprocessorLines = !Line->Tokens.empty(); 2471 ScopedLineState BlockState(*this, SwitchToPreprocessorLines); 2472 assert((LevelDifference >= 0 || 2473 static_cast<unsigned>(-LevelDifference) <= Line->Level) && 2474 "LevelDifference makes Line->Level negative"); 2475 Line->Level += LevelDifference; 2476 // Comments stored before the preprocessor directive need to be output 2477 // before the preprocessor directive, at the same level as the 2478 // preprocessor directive, as we consider them to apply to the directive. 2479 flushComments(isOnNewLine(*FormatTok)); 2480 parsePPDirective(); 2481 } 2482 while (FormatTok->Type == TT_ConflictStart || 2483 FormatTok->Type == TT_ConflictEnd || 2484 FormatTok->Type == TT_ConflictAlternative) { 2485 if (FormatTok->Type == TT_ConflictStart) { 2486 conditionalCompilationStart(/*Unreachable=*/false); 2487 } else if (FormatTok->Type == TT_ConflictAlternative) { 2488 conditionalCompilationAlternative(); 2489 } else if (FormatTok->Type == TT_ConflictEnd) { 2490 conditionalCompilationEnd(); 2491 } 2492 FormatTok = Tokens->getNextToken(); 2493 FormatTok->MustBreakBefore = true; 2494 } 2495 2496 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) && 2497 !Line->InPPDirective) { 2498 continue; 2499 } 2500 2501 if (!FormatTok->Tok.is(tok::comment)) { 2502 distributeComments(Comments, FormatTok); 2503 Comments.clear(); 2504 return; 2505 } 2506 2507 Comments.push_back(FormatTok); 2508 } while (!eof()); 2509 2510 distributeComments(Comments, nullptr); 2511 Comments.clear(); 2512 } 2513 2514 void UnwrappedLineParser::pushToken(FormatToken *Tok) { 2515 Line->Tokens.push_back(UnwrappedLineNode(Tok)); 2516 if (MustBreakBeforeNextToken) { 2517 Line->Tokens.back().Tok->MustBreakBefore = true; 2518 MustBreakBeforeNextToken = false; 2519 } 2520 } 2521 2522 } // end namespace format 2523 } // end namespace clang 2524